{
  "schema": "mithril.mimo-efficiency/v1",
  "conditions": [
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 820.1344427084987,
      "latency_p90_seconds": 3562.819445637501,
      "api_wait_median_seconds": 724.9446559590142,
      "remaining_wall_median_seconds": 95.18978674948448,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 6507.934102708008,
      "episode_latency_median_seconds": 791.9624256874959,
      "cost_per_attempt_usd": 0.826244898,
      "calls_per_attempt": 48.5,
      "metering": {
        "maximum_text_input_proxy_tokens": 105191,
        "maximum_reported_text_prompt_tokens": 104731,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0125640084,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 3.2924155836,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 3.2924155836,
        "physical_text_responses": 194,
        "reported_completion_tokens": 22152,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 194,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 3.304979592,
        "physical_responses": 194,
        "application_cache_hits": 29,
        "application_cache_denominator": 223,
        "provider_cache_hit_responses": 176,
        "provider_cache_metadata_responses": 194,
        "provider_cached_input_tokens": 587520,
        "provider_reported_prompt_tokens": 8136008,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 5639.110055948928,
        "dispatch_wait_seconds": 67.27668341697427,
        "backoff_seconds": 60,
        "observed_callback_seconds": 5784.964299001906
      },
      "categories": {
        "actor": {
          "maximum_text_input_proxy_tokens": 105191,
          "maximum_reported_text_prompt_tokens": 104731,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 3.0639793338,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 3.0639793338,
          "physical_text_responses": 141,
          "reported_completion_tokens": 17713,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 141,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 3.0639793338,
          "physical_responses": 141,
          "application_cache_hits": 23,
          "application_cache_denominator": 164,
          "provider_cache_hit_responses": 128,
          "provider_cache_metadata_responses": 141,
          "provider_cached_input_tokens": 465408,
          "provider_reported_prompt_tokens": 7469761,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 4741.746066365886,
          "dispatch_wait_seconds": 41.8978728329821,
          "backoff_seconds": 45,
          "observed_callback_seconds": 4843.700681795861
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 101435,
          "maximum_reported_text_prompt_tokens": 100521,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.2284362498,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.2284362498,
          "physical_text_responses": 9,
          "reported_completion_tokens": 1905,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 9,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.2284362498,
          "physical_responses": 9,
          "application_cache_hits": 2,
          "application_cache_denominator": 11,
          "provider_cache_hit_responses": 9,
          "provider_cache_metadata_responses": 9,
          "provider_cached_input_tokens": 10368,
          "provider_reported_prompt_tokens": 531613,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 641.4340104169387,
          "dispatch_wait_seconds": 0.9708664990030229,
          "backoff_seconds": 0,
          "observed_callback_seconds": 643.024704041105
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 6118,
          "maximum_reported_text_prompt_tokens": 5997,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.0125640084,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 44,
          "reported_completion_tokens": 2534,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 44,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0125640084,
          "physical_responses": 44,
          "application_cache_hits": 4,
          "application_cache_denominator": 48,
          "provider_cache_hit_responses": 39,
          "provider_cache_metadata_responses": 44,
          "provider_cached_input_tokens": 111744,
          "provider_reported_prompt_tokens": 134634,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 255.9299791661033,
          "dispatch_wait_seconds": 24.40794408498914,
          "backoff_seconds": 15,
          "observed_callback_seconds": 298.23891316493973
        }
      }
    },
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 1147.1746780414978,
      "latency_p90_seconds": 1876.0096806586866,
      "api_wait_median_seconds": 1039.2625193319836,
      "remaining_wall_median_seconds": 71.06654106239148,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 4851.003529874986,
      "episode_latency_median_seconds": 1119.8811749374872,
      "cost_per_attempt_usd": 0.65502481505,
      "calls_per_attempt": 119.25,
      "metering": {
        "maximum_text_input_proxy_tokens": 50785,
        "maximum_reported_text_prompt_tokens": 49690,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.014765403,
        "native_decision_cost_usd": 0.01457276,
        "actor_text_cost_usd": 2.5907610972,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 2.6053338571999998,
        "physical_text_responses": 367,
        "reported_completion_tokens": 70938,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 367,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 2.6200992602,
        "physical_responses": 477,
        "application_cache_hits": 0,
        "application_cache_denominator": 477,
        "provider_cache_hit_responses": 351,
        "provider_cache_metadata_responses": 367,
        "provider_cached_input_tokens": 2607232,
        "provider_reported_prompt_tokens": 8433495,
        "native_decision_responses": 110,
        "provider_error_responses": 0,
        "physical_api_seconds": 4519.466741824988,
        "dispatch_wait_seconds": 73.4193190948863,
        "backoff_seconds": 15,
        "observed_callback_seconds": 4626.135958179686
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 50785,
          "maximum_reported_text_prompt_tokens": 49690,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 1.9632440346,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 1.9632440346,
          "physical_text_responses": 163,
          "reported_completion_tokens": 34224,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 163,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 1.9632440346,
          "physical_responses": 163,
          "application_cache_hits": 0,
          "application_cache_denominator": 163,
          "provider_cache_hit_responses": 157,
          "provider_cache_metadata_responses": 163,
          "provider_cached_input_tokens": 193536,
          "provider_reported_prompt_tokens": 4636691,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 2476.043825783883,
          "dispatch_wait_seconds": 37.443626048014266,
          "backoff_seconds": 15,
          "observed_callback_seconds": 2535.7436914980353
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 18786,
          "maximum_reported_text_prompt_tokens": 18531,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0072093654,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0072093654,
          "physical_text_responses": 2,
          "reported_completion_tokens": 413,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 2,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0072093654,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 2,
          "provider_cache_metadata_responses": 2,
          "provider_cached_input_tokens": 4864,
          "provider_reported_prompt_tokens": 20571,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 17.42495779300225,
          "dispatch_wait_seconds": 0.002916958008427173,
          "backoff_seconds": 0,
          "observed_callback_seconds": 17.470348875009222
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.01457276,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.01457276,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.01457276,
          "physical_responses": 110,
          "application_cache_hits": 0,
          "application_cache_denominator": 110,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 110,
          "provider_error_responses": 0,
          "physical_api_seconds": 116.2365646698745,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 119.72459846694255
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 48985,
          "maximum_reported_text_prompt_tokens": 46673,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.6203076972,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.6203076972,
          "physical_text_responses": 130,
          "reported_completion_tokens": 32092,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 130,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.6203076972,
          "physical_responses": 130,
          "application_cache_hits": 0,
          "application_cache_denominator": 130,
          "provider_cache_hit_responses": 124,
          "provider_cache_metadata_responses": 130,
          "provider_cached_input_tokens": 2196352,
          "provider_reported_prompt_tokens": 3539986,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 1376.516283297009,
          "dispatch_wait_seconds": 27.130025791033404,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1408.874114840699
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 6619,
          "maximum_reported_text_prompt_tokens": 6580,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.014765403,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 72,
          "reported_completion_tokens": 4209,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 72,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.014765403,
          "physical_responses": 72,
          "application_cache_hits": 0,
          "application_cache_denominator": 72,
          "provider_cache_hit_responses": 68,
          "provider_cache_metadata_responses": 72,
          "provider_cached_input_tokens": 212480,
          "provider_reported_prompt_tokens": 236247,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 533.2451102812192,
          "dispatch_wait_seconds": 8.842750297830207,
          "backoff_seconds": 0,
          "observed_callback_seconds": 544.323204499
        }
      }
    },
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 1002.5419530624931,
      "latency_p90_seconds": 2664.225913241794,
      "api_wait_median_seconds": 899.3921287908306,
      "remaining_wall_median_seconds": 103.14982427166251,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 5680.571007000981,
      "episode_latency_median_seconds": 724.1820062289917,
      "cost_per_attempt_usd": 0.84986548645,
      "calls_per_attempt": 143.5,
      "metering": {
        "maximum_text_input_proxy_tokens": 65214,
        "maximum_reported_text_prompt_tokens": 64198,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0167076156,
        "native_decision_cost_usd": 0.018873421,
        "actor_text_cost_usd": 3.3638809092,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 3.3827543302,
        "physical_text_responses": 445,
        "reported_completion_tokens": 90135,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 445,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 3.3994619458,
        "physical_responses": 574,
        "application_cache_hits": 274,
        "application_cache_denominator": 848,
        "provider_cache_hit_responses": 432,
        "provider_cache_metadata_responses": 445,
        "provider_cached_input_tokens": 3245568,
        "provider_reported_prompt_tokens": 10809906,
        "native_decision_responses": 129,
        "provider_error_responses": 0,
        "physical_api_seconds": 5187.153792424389,
        "dispatch_wait_seconds": 64.75411829026416,
        "backoff_seconds": 30,
        "observed_callback_seconds": 5310.0295554229815
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 65214,
          "maximum_reported_text_prompt_tokens": 64198,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 2.4720680526,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 2.4720680526,
          "physical_text_responses": 192,
          "reported_completion_tokens": 30247,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 192,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 2.4720680526,
          "physical_responses": 192,
          "application_cache_hits": 90,
          "application_cache_denominator": 282,
          "provider_cache_hit_responses": 187,
          "provider_cache_metadata_responses": 192,
          "provider_cached_input_tokens": 233216,
          "provider_reported_prompt_tokens": 5853707,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 2229.863593919872,
          "dispatch_wait_seconds": 23.936955585086253,
          "backoff_seconds": 15,
          "observed_callback_seconds": 2278.052230586065
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 9532,
          "maximum_reported_text_prompt_tokens": 9101,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0031006578,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0031006578,
          "physical_text_responses": 1,
          "reported_completion_tokens": 29,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 1,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0031006578,
          "physical_responses": 1,
          "application_cache_hits": 0,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 1,
          "provider_cache_metadata_responses": 1,
          "provider_cached_input_tokens": 2048,
          "provider_reported_prompt_tokens": 9101,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 2.385627082985593,
          "dispatch_wait_seconds": 0.001203457999508828,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2.4001035830006003
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.018873421,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.018873421,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.018873421,
          "physical_responses": 129,
          "application_cache_hits": 64,
          "application_cache_denominator": 193,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 129,
          "provider_error_responses": 0,
          "physical_api_seconds": 173.54467229614966,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 176.9906652495556
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 61831,
          "maximum_reported_text_prompt_tokens": 59120,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.8887121988,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.8887121988,
          "physical_text_responses": 155,
          "reported_completion_tokens": 54907,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 155,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.8887121988,
          "physical_responses": 155,
          "application_cache_hits": 71,
          "application_cache_denominator": 226,
          "provider_cache_hit_responses": 150,
          "provider_cache_metadata_responses": 155,
          "provider_cached_input_tokens": 2619008,
          "provider_reported_prompt_tokens": 4530536,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 2000.0498629116337,
          "dispatch_wait_seconds": 30.740899207012262,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2040.5461163753644
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 10152,
          "maximum_reported_text_prompt_tokens": 10090,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.0167076156,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 97,
          "reported_completion_tokens": 4952,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 97,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0167076156,
          "physical_responses": 97,
          "application_cache_hits": 49,
          "application_cache_denominator": 146,
          "provider_cache_hit_responses": 94,
          "provider_cache_metadata_responses": 97,
          "provider_cached_input_tokens": 391296,
          "provider_reported_prompt_tokens": 416562,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 781.3100362137484,
          "dispatch_wait_seconds": 10.075060040166136,
          "backoff_seconds": 15,
          "observed_callback_seconds": 812.040439628996
        }
      }
    },
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 42.121292291994905,
      "latency_p90_seconds": 204.5283948421042,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 42.121292291994905,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 372.1379151259898,
      "episode_latency_median_seconds": 17.419254145977902,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 105191,
        "maximum_reported_text_prompt_tokens": 104731,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 223,
        "application_cache_denominator": 223,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1.5000414158275817
      },
      "categories": {
        "actor": {
          "maximum_text_input_proxy_tokens": 105191,
          "maximum_reported_text_prompt_tokens": 104731,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 164,
          "application_cache_denominator": 164,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1.2708232049189974
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 101435,
          "maximum_reported_text_prompt_tokens": 100521,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 11,
          "application_cache_denominator": 11,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.05198624904733151
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 6118,
          "maximum_reported_text_prompt_tokens": 5997,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 48,
          "application_cache_denominator": 48,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.17723196186125278
        }
      }
    },
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 39.983900916486164,
      "latency_p90_seconds": 47.25463263309503,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 39.983900916486164,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 150.52832708196365,
      "episode_latency_median_seconds": 20.903287062508753,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 50785,
        "maximum_reported_text_prompt_tokens": 49690,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 477,
        "application_cache_denominator": 477,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1.7230415849480778
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 50785,
          "maximum_reported_text_prompt_tokens": 49690,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 163,
          "application_cache_denominator": 163,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.5691094240173697
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 18786,
          "maximum_reported_text_prompt_tokens": 18531,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.005255291995126754
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 110,
          "application_cache_denominator": 110,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.44052499713143334
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 48985,
          "maximum_reported_text_prompt_tokens": 46673,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 130,
          "application_cache_denominator": 130,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.46482866574660875
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 6619,
          "maximum_reported_text_prompt_tokens": 6580,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 72,
          "application_cache_denominator": 72,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.2433232060575392
        }
      }
    },
    {
      "profile": "reference",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 59.591904083499685,
      "latency_p90_seconds": 264.2056320794946,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 59.591904083499685,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 481.4261366669962,
      "episode_latency_median_seconds": 29.509193707999657,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 65214,
        "maximum_reported_text_prompt_tokens": 64198,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 848,
        "application_cache_denominator": 848,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 5.636966971855145
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 65214,
          "maximum_reported_text_prompt_tokens": 64198,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 282,
          "application_cache_denominator": 282,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1.7605798838485498
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 9532,
          "maximum_reported_text_prompt_tokens": 9101,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 1,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.0017429580038879067
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 193,
          "application_cache_denominator": 193,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1.5345186730846763
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 61831,
          "maximum_reported_text_prompt_tokens": 59120,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 226,
          "application_cache_denominator": 226,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1.4849295449093916
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 10152,
          "maximum_reported_text_prompt_tokens": 10090,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 146,
          "application_cache_denominator": 146,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.8551959120086394
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 3,
      "mean_score": 0.3333333333333333,
      "latency_median_seconds": 82.525155041978,
      "latency_p90_seconds": 102.97224000839633,
      "api_wait_median_seconds": 58.454059542069444,
      "remaining_wall_median_seconds": 18.48087791603757,
      "latency_completed_episodes": 3,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 322.32123524995404,
      "episode_latency_median_seconds": 69.95513858299819,
      "cost_per_attempt_usd": 0.052097019,
      "calls_per_attempt": 5.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 116018,
        "maximum_reported_text_prompt_tokens": 89368,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.208388076,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.208388076,
        "physical_text_responses": 20,
        "reported_completion_tokens": 3951,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 20,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.208388076,
        "physical_responses": 20,
        "application_cache_hits": 0,
        "application_cache_denominator": 20,
        "provider_cache_hit_responses": 16,
        "provider_cache_metadata_responses": 20,
        "provider_cached_input_tokens": 770560,
        "provider_reported_prompt_tokens": 1235334,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 248.46975575110991,
        "dispatch_wait_seconds": 0.3302304589014966,
        "backoff_seconds": 0,
        "observed_callback_seconds": 253.3063598779845
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 116018,
          "maximum_reported_text_prompt_tokens": 89368,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.208388076,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.208388076,
          "physical_text_responses": 20,
          "reported_completion_tokens": 3951,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 20,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.208388076,
          "physical_responses": 20,
          "application_cache_hits": 0,
          "application_cache_denominator": 20,
          "provider_cache_hit_responses": 16,
          "provider_cache_metadata_responses": 20,
          "provider_cached_input_tokens": 770560,
          "provider_reported_prompt_tokens": 1235334,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 248.46975575110991,
          "dispatch_wait_seconds": 0.3302304589014966,
          "backoff_seconds": 0,
          "observed_callback_seconds": 253.3063598779845
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 150.26455737501965,
      "latency_p90_seconds": 257.8728425750946,
      "api_wait_median_seconds": 132.71338452101918,
      "remaining_wall_median_seconds": 17.933213022522978,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 663.6673284580465,
      "episode_latency_median_seconds": 138.76677147949522,
      "cost_per_attempt_usd": 0.09319782765,
      "calls_per_attempt": 6.75,
      "metering": {
        "maximum_text_input_proxy_tokens": 130026,
        "maximum_reported_text_prompt_tokens": 108469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.00031722,
        "actor_text_cost_usd": 0.3724740906,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.3727913106,
        "physical_text_responses": 25,
        "reported_completion_tokens": 8109,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 25,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.3727913106,
        "physical_responses": 27,
        "application_cache_hits": 0,
        "application_cache_denominator": 27,
        "provider_cache_hit_responses": 18,
        "provider_cache_metadata_responses": 25,
        "provider_cached_input_tokens": 768896,
        "provider_reported_prompt_tokens": 1602577,
        "native_decision_responses": 2,
        "provider_error_responses": 0,
        "physical_api_seconds": 592.2765077049844,
        "dispatch_wait_seconds": 0.3459532940178178,
        "backoff_seconds": 0,
        "observed_callback_seconds": 597.518601082993
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.00031722,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.00031722,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.00031722,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 2,
          "provider_error_responses": 0,
          "physical_api_seconds": 2.133372332988074,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2.2118551670282613
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 130026,
          "maximum_reported_text_prompt_tokens": 108469,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.228233799,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.228233799,
          "physical_text_responses": 19,
          "reported_completion_tokens": 3812,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 19,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.228233799,
          "physical_responses": 19,
          "application_cache_hits": 0,
          "application_cache_denominator": 19,
          "provider_cache_hit_responses": 15,
          "provider_cache_metadata_responses": 19,
          "provider_cached_input_tokens": 723840,
          "provider_reported_prompt_tokens": 1234901,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 344.55117012298433,
          "dispatch_wait_seconds": 0.27593737602001056,
          "backoff_seconds": 0,
          "observed_callback_seconds": 348.49279162299354
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 109184,
          "maximum_reported_text_prompt_tokens": 87425,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.095669217,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.095669217,
          "physical_text_responses": 4,
          "reported_completion_tokens": 3012,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 4,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.095669217,
          "physical_responses": 4,
          "application_cache_hits": 0,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 2,
          "provider_cache_metadata_responses": 4,
          "provider_cached_input_tokens": 30720,
          "provider_reported_prompt_tokens": 244371,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 208.7506594590377,
          "dispatch_wait_seconds": 0.0453681249928195,
          "backoff_seconds": 0,
          "observed_callback_seconds": 209.58340183398104
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 88828,
          "maximum_reported_text_prompt_tokens": 73310,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0485710746,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0485710746,
          "physical_text_responses": 2,
          "reported_completion_tokens": 1285,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 2,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0485710746,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 1,
          "provider_cache_metadata_responses": 2,
          "provider_cached_input_tokens": 14336,
          "provider_reported_prompt_tokens": 123305,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 36.841305789974285,
          "dispatch_wait_seconds": 0.02464779300498776,
          "backoff_seconds": 0,
          "observed_callback_seconds": 37.23055245899013
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.5,
      "latency_median_seconds": 89.4540095830016,
      "latency_p90_seconds": 157.19887677040828,
      "api_wait_median_seconds": 77.08641247748164,
      "remaining_wall_median_seconds": 16.149687813522178,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 453.1567095410137,
      "episode_latency_median_seconds": 79.99844016649877,
      "cost_per_attempt_usd": 0.08654612025,
      "calls_per_attempt": 6.5,
      "metering": {
        "maximum_text_input_proxy_tokens": 99324,
        "maximum_reported_text_prompt_tokens": 82536,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.346184481,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.346184481,
        "physical_text_responses": 26,
        "reported_completion_tokens": 6009,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 26,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.346184481,
        "physical_responses": 26,
        "application_cache_hits": 0,
        "application_cache_denominator": 26,
        "provider_cache_hit_responses": 19,
        "provider_cache_metadata_responses": 26,
        "provider_cached_input_tokens": 799360,
        "provider_reported_prompt_tokens": 1576553,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 385.47764024388744,
        "dispatch_wait_seconds": 0.2630013760353904,
        "backoff_seconds": 0,
        "observed_callback_seconds": 389.30950833304087
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 99324,
          "maximum_reported_text_prompt_tokens": 82536,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.1550698944,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.1550698944,
          "physical_text_responses": 18,
          "reported_completion_tokens": 3237,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 18,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.1550698944,
          "physical_responses": 18,
          "application_cache_hits": 0,
          "application_cache_denominator": 18,
          "provider_cache_hit_responses": 16,
          "provider_cache_metadata_responses": 18,
          "provider_cached_input_tokens": 754304,
          "provider_reported_prompt_tokens": 1098070,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 212.06464966590283,
          "dispatch_wait_seconds": 0.2180407070554793,
          "backoff_seconds": 0,
          "observed_callback_seconds": 214.86469449903234
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 80044,
          "maximum_reported_text_prompt_tokens": 62229,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.091261797,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.091261797,
          "physical_text_responses": 4,
          "reported_completion_tokens": 1135,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 4,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.091261797,
          "physical_responses": 4,
          "application_cache_hits": 0,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 2,
          "provider_cache_metadata_responses": 4,
          "provider_cached_input_tokens": 30720,
          "provider_reported_prompt_tokens": 237993,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 77.0796147059882,
          "dispatch_wait_seconds": 0.01844091698876582,
          "backoff_seconds": 0,
          "observed_callback_seconds": 77.54164208302973
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 80683,
          "maximum_reported_text_prompt_tokens": 63381,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.09985278959999999,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.09985278959999999,
          "physical_text_responses": 4,
          "reported_completion_tokens": 1637,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 4,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.09985278959999999,
          "physical_responses": 4,
          "application_cache_hits": 0,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 1,
          "provider_cache_metadata_responses": 4,
          "provider_cached_input_tokens": 14336,
          "provider_reported_prompt_tokens": 240490,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 96.3333758719964,
          "dispatch_wait_seconds": 0.026519751991145313,
          "backoff_seconds": 0,
          "observed_callback_seconds": 96.9031717509788
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 3,
      "mean_score": 0.3333333333333333,
      "latency_median_seconds": 1.8242214169877116,
      "latency_p90_seconds": 2.866977650584886,
      "api_wait_median_seconds": 0,
      "remaining_wall_median_seconds": 1.8242214169877116,
      "latency_completed_episodes": 3,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 7.7602629179891665,
      "episode_latency_median_seconds": 1.7584796250157524,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 116018,
        "maximum_reported_text_prompt_tokens": 89368,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 20,
        "application_cache_denominator": 20,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.44813750105095096
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 116018,
          "maximum_reported_text_prompt_tokens": 89368,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 20,
          "application_cache_denominator": 20,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.44813750105095096
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 4.687698625493795,
      "latency_p90_seconds": 5.735591771395411,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 4.687698625493795,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 16.231962959980592,
      "episode_latency_median_seconds": 4.672068395491806,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 130026,
        "maximum_reported_text_prompt_tokens": 108469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 27,
        "application_cache_denominator": 27,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.5256445850245655
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.018576291971839964
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 130026,
          "maximum_reported_text_prompt_tokens": 108469,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 19,
          "application_cache_denominator": 19,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.46127483405871317
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 109184,
          "maximum_reported_text_prompt_tokens": 87425,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 4,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.036097166972467676
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 88828,
          "maximum_reported_text_prompt_tokens": 73310,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.00969629202154465
        }
      }
    },
    {
      "profile": "reference",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.5,
      "latency_median_seconds": 1.8897731454926543,
      "latency_p90_seconds": 3.1566973124048676,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 1.8897731454926543,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 9.013070332992356,
      "episode_latency_median_seconds": 1.8815155210177181,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 99324,
        "maximum_reported_text_prompt_tokens": 82536,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 26,
        "application_cache_denominator": 26,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.18827254307689145
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 99324,
          "maximum_reported_text_prompt_tokens": 82536,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 18,
          "application_cache_denominator": 18,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.16461733303731307
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 80044,
          "maximum_reported_text_prompt_tokens": 62229,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 4,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.01924250103184022
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 80683,
          "maximum_reported_text_prompt_tokens": 63381,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 4,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.004412709007738158
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 454.96934816600697,
      "latency_p90_seconds": 991.0771682004881,
      "api_wait_median_seconds": 410.9866863944917,
      "remaining_wall_median_seconds": 53.01746227151307,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 2203.52665616601,
      "episode_latency_median_seconds": 434.4987659579929,
      "cost_per_attempt_usd": 0.30218060955,
      "calls_per_attempt": 49.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 58729,
        "maximum_reported_text_prompt_tokens": 57812,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0085508238,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 1.2001716144,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 1.2001716143999999,
        "physical_text_responses": 196,
        "reported_completion_tokens": 16825,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 196,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 1.2087224382,
        "physical_responses": 196,
        "application_cache_hits": 0,
        "application_cache_denominator": 196,
        "provider_cache_hit_responses": 194,
        "provider_cache_metadata_responses": 196,
        "provider_cached_input_tokens": 2778112,
        "provider_reported_prompt_tokens": 5500143,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 1942.5700824219966,
        "dispatch_wait_seconds": 14.898216498026159,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1965.7575722899055
      },
      "categories": {
        "actor": {
          "maximum_text_input_proxy_tokens": 58729,
          "maximum_reported_text_prompt_tokens": 57812,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 1.092648912,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 1.092648912,
          "physical_text_responses": 149,
          "reported_completion_tokens": 14226,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 149,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 1.092648912,
          "physical_responses": 149,
          "application_cache_hits": 0,
          "application_cache_denominator": 149,
          "provider_cache_hit_responses": 148,
          "provider_cache_metadata_responses": 149,
          "provider_cached_input_tokens": 2657920,
          "provider_reported_prompt_tokens": 5119308,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 1630.9210163349344,
          "dispatch_wait_seconds": 9.046868292003637,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1645.817468792753
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 48187,
          "maximum_reported_text_prompt_tokens": 46929,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.10752270239999999,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.10752270239999999,
          "physical_text_responses": 6,
          "reported_completion_tokens": 696,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 6,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.10752270239999999,
          "physical_responses": 6,
          "application_cache_hits": 0,
          "application_cache_denominator": 6,
          "provider_cache_hit_responses": 6,
          "provider_cache_metadata_responses": 6,
          "provider_cached_input_tokens": 9984,
          "provider_reported_prompt_tokens": 255688,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 174.84722804097692,
          "dispatch_wait_seconds": 0.014898876019287854,
          "backoff_seconds": 0,
          "observed_callback_seconds": 175.30458733299747
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 7192,
          "maximum_reported_text_prompt_tokens": 7075,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.0085508238,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 41,
          "reported_completion_tokens": 1903,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 41,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0085508238,
          "physical_responses": 41,
          "application_cache_hits": 0,
          "application_cache_denominator": 41,
          "provider_cache_hit_responses": 40,
          "provider_cache_metadata_responses": 41,
          "provider_cached_input_tokens": 110208,
          "provider_reported_prompt_tokens": 125147,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 136.80183804608532,
          "dispatch_wait_seconds": 5.836449330003234,
          "backoff_seconds": 0,
          "observed_callback_seconds": 144.63551616415498
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 918.3535884380108,
      "latency_p90_seconds": 2239.9864575707998,
      "api_wait_median_seconds": 829.196659868976,
      "remaining_wall_median_seconds": 89.15692856903479,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 6,
      "total_attempted_wall_seconds": 4843.111666376004,
      "episode_latency_median_seconds": 790.1741025624942,
      "cost_per_attempt_usd": 0.3762208421,
      "calls_per_attempt": 137.5,
      "metering": {
        "maximum_text_input_proxy_tokens": 55493,
        "maximum_reported_text_prompt_tokens": 54055,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.008122551,
        "native_decision_cost_usd": 0.03057662,
        "actor_text_cost_usd": 1.4661841974,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 1.4967608174,
        "physical_text_responses": 320,
        "reported_completion_tokens": 56089,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 320,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 1.5048833684,
        "physical_responses": 550,
        "application_cache_hits": 65,
        "application_cache_denominator": 615,
        "provider_cache_hit_responses": 310,
        "provider_cache_metadata_responses": 320,
        "provider_cached_input_tokens": 4818944,
        "provider_reported_prompt_tokens": 8056096,
        "native_decision_responses": 230,
        "provider_error_responses": 0,
        "physical_api_seconds": 4463.519989065127,
        "dispatch_wait_seconds": 25.058096004766412,
        "backoff_seconds": 0,
        "observed_callback_seconds": 4505.618258496077
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 55493,
          "maximum_reported_text_prompt_tokens": 54055,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.8639641884,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.8639641884,
          "physical_text_responses": 150,
          "reported_completion_tokens": 26626,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 150,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.8639641884,
          "physical_responses": 150,
          "application_cache_hits": 19,
          "application_cache_denominator": 169,
          "provider_cache_hit_responses": 148,
          "provider_cache_metadata_responses": 150,
          "provider_cached_input_tokens": 2655744,
          "provider_reported_prompt_tokens": 4566638,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 1997.63664207299,
          "dispatch_wait_seconds": 11.493981954903575,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2015.1075941290765
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 45875,
          "maximum_reported_text_prompt_tokens": 44920,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.005692237199999999,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.005692237199999999,
          "physical_text_responses": 7,
          "reported_completion_tokens": 540,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 7,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.005692237199999999,
          "physical_responses": 7,
          "application_cache_hits": 3,
          "application_cache_denominator": 10,
          "provider_cache_hit_responses": 7,
          "provider_cache_metadata_responses": 7,
          "provider_cached_input_tokens": 173952,
          "provider_reported_prompt_tokens": 184518,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 35.241575001040474,
          "dispatch_wait_seconds": 1.2935545409563929,
          "backoff_seconds": 0,
          "observed_callback_seconds": 36.728986709029414
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.011936800000000001,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.011936800000000001,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.011936800000000001,
          "physical_responses": 88,
          "application_cache_hits": 11,
          "application_cache_denominator": 99,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 88,
          "provider_error_responses": 0,
          "physical_api_seconds": 85.0180261649657,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 86.92557383619715
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 53578,
          "maximum_reported_text_prompt_tokens": 50794,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.5965277718,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.5965277718,
          "physical_text_responses": 110,
          "reported_completion_tokens": 26665,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 110,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.5965277718,
          "physical_responses": 110,
          "application_cache_hits": 12,
          "application_cache_denominator": 122,
          "provider_cache_hit_responses": 103,
          "provider_cache_metadata_responses": 110,
          "provider_cached_input_tokens": 1865088,
          "provider_reported_prompt_tokens": 3167651,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 2025.2108897031285,
          "dispatch_wait_seconds": 5.480715712008532,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2034.79529557575
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.01863982,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.01863982,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.01863982,
          "physical_responses": 142,
          "application_cache_hits": 17,
          "application_cache_denominator": 159,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 142,
          "provider_error_responses": 0,
          "physical_api_seconds": 140.34183250193018,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 143.850205579045
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 4719,
          "maximum_reported_text_prompt_tokens": 4567,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.008122551,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 53,
          "reported_completion_tokens": 2258,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 53,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.008122551,
          "physical_responses": 53,
          "application_cache_hits": 3,
          "application_cache_denominator": 56,
          "provider_cache_hit_responses": 52,
          "provider_cache_metadata_responses": 53,
          "provider_cached_input_tokens": 124160,
          "provider_reported_prompt_tokens": 137289,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 180.07102362107253,
          "dispatch_wait_seconds": 6.789843796897912,
          "backoff_seconds": 0,
          "observed_callback_seconds": 188.21060266697896
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "primary",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 701.7196970415098,
      "latency_p90_seconds": 1447.7351951324235,
      "api_wait_median_seconds": 617.2100782716298,
      "remaining_wall_median_seconds": 86.56813733030867,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 6,
      "total_attempted_wall_seconds": 3396.2462223740295,
      "episode_latency_median_seconds": 454.289333812485,
      "cost_per_attempt_usd": 0.28884523545,
      "calls_per_attempt": 112.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 53895,
        "maximum_reported_text_prompt_tokens": 52417,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0099682158,
        "native_decision_cost_usd": 0.022785933,
        "actor_text_cost_usd": 1.122626793,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 1.145412726,
        "physical_text_responses": 292,
        "reported_completion_tokens": 49265,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 292,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 1.1553809418,
        "physical_responses": 448,
        "application_cache_hits": 119,
        "application_cache_denominator": 567,
        "provider_cache_hit_responses": 286,
        "provider_cache_metadata_responses": 292,
        "provider_cached_input_tokens": 4542208,
        "provider_reported_prompt_tokens": 7009754,
        "native_decision_responses": 156,
        "provider_error_responses": 0,
        "physical_api_seconds": 3012.078233661363,
        "dispatch_wait_seconds": 27.759985883691115,
        "backoff_seconds": 0,
        "observed_callback_seconds": 3055.2403478702763
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 53895,
          "maximum_reported_text_prompt_tokens": 52417,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.733315245,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.733315245,
          "physical_text_responses": 140,
          "reported_completion_tokens": 22808,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 140,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.733315245,
          "physical_responses": 140,
          "application_cache_hits": 34,
          "application_cache_denominator": 174,
          "provider_cache_hit_responses": 139,
          "provider_cache_metadata_responses": 140,
          "provider_cached_input_tokens": 2566400,
          "provider_reported_prompt_tokens": 4185327,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 1494.9023365820758,
          "dispatch_wait_seconds": 11.06247496290598,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1511.096239041246
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 26252,
          "maximum_reported_text_prompt_tokens": 25748,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0014992278,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0014992278,
          "physical_text_responses": 4,
          "reported_completion_tokens": 565,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 4,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0014992278,
          "physical_responses": 4,
          "application_cache_hits": 2,
          "application_cache_denominator": 6,
          "provider_cache_hit_responses": 4,
          "provider_cache_metadata_responses": 4,
          "provider_cached_input_tokens": 85248,
          "provider_reported_prompt_tokens": 86859,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 31.641914248961257,
          "dispatch_wait_seconds": 1.2090420839958824,
          "backoff_seconds": 0,
          "observed_callback_seconds": 32.92721733401413
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.010786735,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.010786735,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.010786735,
          "physical_responses": 74,
          "application_cache_hits": 18,
          "application_cache_denominator": 92,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 74,
          "provider_error_responses": 0,
          "physical_api_seconds": 84.07526808095281,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 86.176918375073
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 51668,
          "maximum_reported_text_prompt_tokens": 49131,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.3878123202,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.3878123202,
          "physical_text_responses": 94,
          "reported_completion_tokens": 22953,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 94,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.3878123202,
          "physical_responses": 94,
          "application_cache_hits": 18,
          "application_cache_denominator": 112,
          "provider_cache_hit_responses": 92,
          "provider_cache_metadata_responses": 94,
          "provider_cached_input_tokens": 1775232,
          "provider_reported_prompt_tokens": 2606157,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 1123.8770913319895,
          "dispatch_wait_seconds": 8.968819752917625,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1136.9550962508074
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.011999197999999999,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.011999197999999999,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.011999197999999999,
          "physical_responses": 82,
          "application_cache_hits": 34,
          "application_cache_denominator": 116,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 82,
          "provider_error_responses": 0,
          "physical_api_seconds": 104.57579283817904,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 107.19718332806951
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 4358,
          "maximum_reported_text_prompt_tokens": 4259,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0.0099682158,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0,
          "physical_text_responses": 54,
          "reported_completion_tokens": 2939,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 54,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0099682158,
          "physical_responses": 54,
          "application_cache_hits": 13,
          "application_cache_denominator": 67,
          "provider_cache_hit_responses": 51,
          "provider_cache_metadata_responses": 54,
          "provider_cached_input_tokens": 115328,
          "provider_reported_prompt_tokens": 131411,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 173.00583057920448,
          "dispatch_wait_seconds": 6.519649083871627,
          "backoff_seconds": 0,
          "observed_callback_seconds": 180.88769354106626
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 35.683247187495,
      "latency_p90_seconds": 89.66688750898467,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 35.683247187495,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 5,
      "total_attempted_wall_seconds": 210.42244433396263,
      "episode_latency_median_seconds": 13.643351583508775,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 58729,
        "maximum_reported_text_prompt_tokens": 57812,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 196,
        "application_cache_denominator": 196,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.9318176987871993
      },
      "categories": {
        "actor": {
          "maximum_text_input_proxy_tokens": 58729,
          "maximum_reported_text_prompt_tokens": 57812,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 149,
          "application_cache_denominator": 149,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.7669052009296138
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 48187,
          "maximum_reported_text_prompt_tokens": 46929,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 6,
          "application_cache_denominator": 6,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.01957354199839756
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 7192,
          "maximum_reported_text_prompt_tokens": 7075,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 41,
          "application_cache_denominator": 41,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.145338955859188
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 77.02185693749925,
      "latency_p90_seconds": 107.85701044110465,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 77.02185693749925,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 6,
      "total_attempted_wall_seconds": 300.1938318320317,
      "episode_latency_median_seconds": 28.65214672900038,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 55493,
        "maximum_reported_text_prompt_tokens": 54055,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 615,
        "application_cache_denominator": 615,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 2.1203582261805423
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 55493,
          "maximum_reported_text_prompt_tokens": 54055,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 169,
          "application_cache_denominator": 169,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.4921573020401411
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 45875,
          "maximum_reported_text_prompt_tokens": 44920,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 10,
          "application_cache_denominator": 10,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.03575025105965324
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 99,
          "application_cache_denominator": 99,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.3652275890635792
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 53578,
          "maximum_reported_text_prompt_tokens": 50794,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 122,
          "application_cache_denominator": 122,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.45152687703375705
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 159,
          "application_cache_denominator": 159,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.6079671689076349
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 4719,
          "maximum_reported_text_prompt_tokens": 4567,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 56,
          "application_cache_denominator": 56,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.16772903807577677
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "banking_knowledge",
      "phase": "warm",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.0,
      "latency_median_seconds": 62.23326622949389,
      "latency_p90_seconds": 101.60457020450559,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 62.23326622949389,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 6,
      "total_attempted_wall_seconds": 260.1935522920103,
      "episode_latency_median_seconds": 26.758683041494805,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 53895,
        "maximum_reported_text_prompt_tokens": 52417,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 567,
        "application_cache_denominator": 567,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1.9749248523148708
      },
      "categories": {
        "actor-joint-candidates": {
          "maximum_text_input_proxy_tokens": 53895,
          "maximum_reported_text_prompt_tokens": 52417,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 174,
          "application_cache_denominator": 174,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.629092080227565
        },
        "actor-recovery": {
          "maximum_text_input_proxy_tokens": 26252,
          "maximum_reported_text_prompt_tokens": 25748,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 6,
          "application_cache_denominator": 6,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.016174748947378248
        },
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 92,
          "application_cache_denominator": 92,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.29078386988840066
        },
        "review-and-revision": {
          "maximum_text_input_proxy_tokens": 51668,
          "maximum_reported_text_prompt_tokens": 49131,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 112,
          "application_cache_denominator": 112,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.4237331190670375
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 116,
          "application_cache_denominator": 116,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.3716210701386444
        },
        "user-simulator": {
          "maximum_text_input_proxy_tokens": 4358,
          "maximum_reported_text_prompt_tokens": 4259,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 67,
          "application_cache_denominator": 67,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.24351996404584497
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.25,
      "latency_median_seconds": 112.01362068750313,
      "latency_p90_seconds": 159.5908057669003,
      "api_wait_median_seconds": 94.9553723760182,
      "remaining_wall_median_seconds": 24.00440270701074,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 484.0392724580306,
      "episode_latency_median_seconds": 98.93953931250144,
      "cost_per_attempt_usd": 0.0531723951,
      "calls_per_attempt": 5.25,
      "metering": {
        "maximum_text_input_proxy_tokens": 124779,
        "maximum_reported_text_prompt_tokens": 104642,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.2126895804,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.2126895804,
        "physical_text_responses": 21,
        "reported_completion_tokens": 3679,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 21,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.2126895804,
        "physical_responses": 21,
        "application_cache_hits": 0,
        "application_cache_denominator": 21,
        "provider_cache_hit_responses": 18,
        "provider_cache_metadata_responses": 21,
        "provider_cached_input_tokens": 968064,
        "provider_reported_prompt_tokens": 1441636,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 371.94986071105814,
        "dispatch_wait_seconds": 0.3067695829377044,
        "backoff_seconds": 0,
        "observed_callback_seconds": 376.8610252070939
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 124779,
          "maximum_reported_text_prompt_tokens": 104642,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.2126895804,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.2126895804,
          "physical_text_responses": 21,
          "reported_completion_tokens": 3679,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 21,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.2126895804,
          "physical_responses": 21,
          "application_cache_hits": 0,
          "application_cache_denominator": 21,
          "provider_cache_hit_responses": 18,
          "provider_cache_metadata_responses": 21,
          "provider_cached_input_tokens": 968064,
          "provider_reported_prompt_tokens": 1441636,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 371.94986071105814,
          "dispatch_wait_seconds": 0.3067695829377044,
          "backoff_seconds": 0,
          "observed_callback_seconds": 376.8610252070939
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.5,
      "latency_median_seconds": 88.35388874998898,
      "latency_p90_seconds": 99.40792324129725,
      "api_wait_median_seconds": 63.953042456530966,
      "remaining_wall_median_seconds": 32.31397562545317,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 356.00074995699106,
      "episode_latency_median_seconds": 72.86722868749348,
      "cost_per_attempt_usd": 0.04819704965,
      "calls_per_attempt": 9.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 74611,
        "maximum_reported_text_prompt_tokens": 68569,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.00275504,
        "actor_text_cost_usd": 0.1900331586,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.1927881986,
        "physical_text_responses": 18,
        "reported_completion_tokens": 5138,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 18,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.1927881986,
        "physical_responses": 36,
        "application_cache_hits": 0,
        "application_cache_denominator": 36,
        "provider_cache_hit_responses": 12,
        "provider_cache_metadata_responses": 18,
        "provider_cached_input_tokens": 405376,
        "provider_reported_prompt_tokens": 828603,
        "native_decision_responses": 18,
        "provider_error_responses": 0,
        "physical_api_seconds": 236.6993843691016,
        "dispatch_wait_seconds": 0.13697653997223824,
        "backoff_seconds": 0,
        "observed_callback_seconds": 242.35854725111858
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.00028228,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.00028228,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.00028228,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 2,
          "provider_error_responses": 0,
          "physical_api_seconds": 1.9420100819843356,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 2.07349866599543
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 74611,
          "maximum_reported_text_prompt_tokens": 68569,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.113951712,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.113951712,
          "physical_text_responses": 12,
          "reported_completion_tokens": 2402,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 12,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.113951712,
          "physical_responses": 12,
          "application_cache_hits": 0,
          "application_cache_denominator": 12,
          "provider_cache_hit_responses": 9,
          "provider_cache_metadata_responses": 12,
          "provider_cached_input_tokens": 360320,
          "provider_reported_prompt_tokens": 614492,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 121.96346716905828,
          "dispatch_wait_seconds": 0.07396228896686807,
          "backoff_seconds": 0,
          "observed_callback_seconds": 124.76828925305745
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 53149,
          "maximum_reported_text_prompt_tokens": 46474,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.049299957,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.049299957,
          "physical_text_responses": 4,
          "reported_completion_tokens": 2000,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 4,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.049299957,
          "physical_responses": 4,
          "application_cache_hits": 0,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 2,
          "provider_cache_metadata_responses": 4,
          "provider_cached_input_tokens": 30720,
          "provider_reported_prompt_tokens": 139799,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 67.78819725001813,
          "dispatch_wait_seconds": 0.05950566698447801,
          "backoff_seconds": 0,
          "observed_callback_seconds": 68.29864570798236
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 53507,
          "maximum_reported_text_prompt_tokens": 46828,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0267814896,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0267814896,
          "physical_text_responses": 2,
          "reported_completion_tokens": 736,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 2,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0267814896,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 1,
          "provider_cache_metadata_responses": 2,
          "provider_cached_input_tokens": 14336,
          "provider_reported_prompt_tokens": 74312,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 20.491719747980824,
          "dispatch_wait_seconds": 0.003508584020892158,
          "backoff_seconds": 0,
          "observed_callback_seconds": 20.760141958016902
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.00247276,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.00247276,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.00247276,
          "physical_responses": 16,
          "application_cache_hits": 0,
          "application_cache_denominator": 16,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 16,
          "provider_error_responses": 0,
          "physical_api_seconds": 24.513990120060043,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 26.45797166606644
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "primary",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 2,
      "mean_score": 0.0,
      "latency_median_seconds": 79.0887275210116,
      "latency_p90_seconds": 80.54145683781243,
      "api_wait_median_seconds": 55.804185228960705,
      "remaining_wall_median_seconds": 23.2845422920509,
      "latency_completed_episodes": 2,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 305.66674133299966,
      "episode_latency_median_seconds": 65.85301116701157,
      "cost_per_attempt_usd": 0.03729586895,
      "calls_per_attempt": 7.25,
      "metering": {
        "maximum_text_input_proxy_tokens": 66349,
        "maximum_reported_text_prompt_tokens": 60644,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.002949674,
        "actor_text_cost_usd": 0.1462338018,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.1491834758,
        "physical_text_responses": 12,
        "reported_completion_tokens": 4987,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 12,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.1491834758,
        "physical_responses": 29,
        "application_cache_hits": 0,
        "application_cache_denominator": 29,
        "provider_cache_hit_responses": 9,
        "provider_cache_metadata_responses": 12,
        "provider_cached_input_tokens": 297088,
        "provider_reported_prompt_tokens": 620825,
        "native_decision_responses": 17,
        "provider_error_responses": 0,
        "physical_api_seconds": 229.19135462492704,
        "dispatch_wait_seconds": 0.259303622005973,
        "backoff_seconds": 0,
        "observed_callback_seconds": 235.97482074893196
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.000136629,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.000136629,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.000136629,
          "physical_responses": 1,
          "application_cache_hits": 0,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 1,
          "provider_error_responses": 0,
          "physical_api_seconds": 0.9368131659866776,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 1.2142728329927195
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 66349,
          "maximum_reported_text_prompt_tokens": 60644,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.110482716,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.110482716,
          "physical_text_responses": 9,
          "reported_completion_tokens": 3614,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 9,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.110482716,
          "physical_responses": 9,
          "application_cache_hits": 0,
          "application_cache_denominator": 9,
          "provider_cache_hit_responses": 6,
          "provider_cache_metadata_responses": 9,
          "provider_cached_input_tokens": 268160,
          "provider_reported_prompt_tokens": 512696,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 149.21206812694436,
          "dispatch_wait_seconds": 0.2410925390140619,
          "backoff_seconds": 0,
          "observed_callback_seconds": 151.57116158292047
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 42693,
          "maximum_reported_text_prompt_tokens": 37249,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0180782892,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0180782892,
          "physical_text_responses": 2,
          "reported_completion_tokens": 731,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 2,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0180782892,
          "physical_responses": 2,
          "application_cache_hits": 0,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 2,
          "provider_cache_metadata_responses": 2,
          "provider_cached_input_tokens": 28672,
          "provider_reported_prompt_tokens": 68532,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 39.18330287502613,
          "dispatch_wait_seconds": 0.0070288749993778765,
          "backoff_seconds": 0,
          "observed_callback_seconds": 39.393849667016184
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 45250,
          "maximum_reported_text_prompt_tokens": 39597,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0.0176727966,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.0176727966,
          "physical_text_responses": 1,
          "reported_completion_tokens": 642,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 1,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.0176727966,
          "physical_responses": 1,
          "application_cache_hits": 0,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 1,
          "provider_cache_metadata_responses": 1,
          "provider_cached_input_tokens": 256,
          "provider_reported_prompt_tokens": 39597,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 17.342829417000758,
          "dispatch_wait_seconds": 0.011182207992533222,
          "backoff_seconds": 0,
          "observed_callback_seconds": 17.44606700001168
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0.002813045,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0.002813045,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.002813045,
          "physical_responses": 16,
          "application_cache_hits": 0,
          "application_cache_denominator": 16,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 16,
          "provider_error_responses": 0,
          "physical_api_seconds": 22.51634103996912,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 26.34946966599091
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "actor-only",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.25,
      "latency_median_seconds": 1.7366541664960096,
      "latency_p90_seconds": 3.047643649901147,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 1.7366541664960096,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 7.835662916011643,
      "episode_latency_median_seconds": 1.704102874995442,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 124779,
        "maximum_reported_text_prompt_tokens": 104642,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 21,
        "application_cache_denominator": 21,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.1316028320870828
      },
      "categories": {
        "it-actor": {
          "maximum_text_input_proxy_tokens": 124779,
          "maximum_reported_text_prompt_tokens": 104642,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 21,
          "application_cache_denominator": 21,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.1316028320870828
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "planned": 4,
      "attempted": 4,
      "scored": 4,
      "mean_score": 0.5,
      "latency_median_seconds": 4.034787354510627,
      "latency_p90_seconds": 6.524598683114164,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 4.034787354510627,
      "latency_completed_episodes": 4,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 16.768864042067435,
      "episode_latency_median_seconds": 4.006302145498921,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 74611,
        "maximum_reported_text_prompt_tokens": 68569,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 36,
        "application_cache_denominator": 36,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.16495375195518136
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.002848457981599495
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 74611,
          "maximum_reported_text_prompt_tokens": 68569,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 12,
          "application_cache_denominator": 12,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.05417587497504428
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 53149,
          "maximum_reported_text_prompt_tokens": 46474,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 4,
          "application_cache_denominator": 4,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.021842250047484413
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 53507,
          "maximum_reported_text_prompt_tokens": 46828,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.00411391700617969
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 16,
          "application_cache_denominator": 16,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.08197325194487348
        }
      }
    },
    {
      "profile": "ontology",
      "domain": "ITBench",
      "phase": "warm",
      "arm": "cloudflare/clef-flash",
      "planned": 4,
      "attempted": 4,
      "scored": 2,
      "mean_score": 0.0,
      "latency_median_seconds": 12.168910770997172,
      "latency_p90_seconds": 13.99243848779588,
      "api_wait_median_seconds": 0.0,
      "remaining_wall_median_seconds": 12.168910770997172,
      "latency_completed_episodes": 2,
      "actual_original_attempts": 4,
      "total_attempted_wall_seconds": 34.25321608298691,
      "episode_latency_median_seconds": 11.961229916996672,
      "cost_per_attempt_usd": 0.0,
      "calls_per_attempt": 0.0,
      "metering": {
        "maximum_text_input_proxy_tokens": 66349,
        "maximum_reported_text_prompt_tokens": 60644,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 29,
        "application_cache_denominator": 29,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.23114104196429253
      },
      "categories": {
        "decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 1,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.006713999988278374
        },
        "it-actor": {
          "maximum_text_input_proxy_tokens": 66349,
          "maximum_reported_text_prompt_tokens": 60644,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 9,
          "application_cache_denominator": 9,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.0690713340009097
        },
        "it-critique": {
          "maximum_text_input_proxy_tokens": 42693,
          "maximum_reported_text_prompt_tokens": 37249,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 2,
          "application_cache_denominator": 2,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.04571516698342748
        },
        "it-revise-alternatives": {
          "maximum_text_input_proxy_tokens": 45250,
          "maximum_reported_text_prompt_tokens": 39597,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 1,
          "application_cache_denominator": 1,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.01316249999217689
        },
        "source-control-decision": {
          "maximum_text_input_proxy_tokens": 0,
          "maximum_reported_text_prompt_tokens": 0,
          "text_input_proxy_limit": 262144,
          "simulator_cost_usd": 0,
          "native_decision_cost_usd": 0,
          "actor_text_cost_usd": 0,
          "other_settled_cost_usd": 0,
          "agent_side_cost_usd": 0,
          "physical_text_responses": 0,
          "reported_completion_tokens": 0,
          "reported_reasoning_tokens": 0,
          "reasoning_metadata_responses": 0,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0,
          "physical_responses": 0,
          "application_cache_hits": 16,
          "application_cache_denominator": 16,
          "provider_cache_hit_responses": 0,
          "provider_cache_metadata_responses": 0,
          "provider_cached_input_tokens": 0,
          "provider_reported_prompt_tokens": 0,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 0,
          "dispatch_wait_seconds": 0,
          "backoff_seconds": 0,
          "observed_callback_seconds": 0.09647804099950008
        }
      }
    }
  ],
  "paired_ontology": [
    {
      "domain": "banking_knowledge",
      "arm": "actor-only",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "reference_mean_score": 0.0,
      "ontology_mean_score": 0.0,
      "reference_latency_median_seconds": 820.1344427084987,
      "ontology_latency_median_seconds": 454.96934816600697,
      "reference_cost_usd": 3.304979592,
      "ontology_cost_usd": 1.2087224382
    },
    {
      "domain": "banking_knowledge",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "reference_mean_score": 0.0,
      "ontology_mean_score": 0.0,
      "reference_latency_median_seconds": 1147.1746780414978,
      "ontology_latency_median_seconds": 918.3535884380108,
      "reference_cost_usd": 2.6200992602,
      "ontology_cost_usd": 1.5048833684
    },
    {
      "domain": "banking_knowledge",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "reference_mean_score": 0.0,
      "ontology_mean_score": 0.0,
      "reference_latency_median_seconds": 1002.5419530624931,
      "ontology_latency_median_seconds": 701.7196970415098,
      "reference_cost_usd": 3.3994619458,
      "ontology_cost_usd": 1.1553809418
    },
    {
      "domain": "ITBench",
      "arm": "actor-only",
      "complete_pairs": 3,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33"
      ],
      "reference_mean_score": 0.3333333333333333,
      "ontology_mean_score": 0.3333333333333333,
      "reference_latency_median_seconds": 82.525155041978,
      "ontology_latency_median_seconds": 127.78799650000292,
      "reference_cost_usd": 0.1140022608,
      "ontology_cost_usd": 0.1683665532
    },
    {
      "domain": "ITBench",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33",
        "Scenario-4"
      ],
      "reference_mean_score": 0.0,
      "ontology_mean_score": 0.5,
      "reference_latency_median_seconds": 150.26455737501965,
      "ontology_latency_median_seconds": 88.35388874998898,
      "reference_cost_usd": 0.3727913106,
      "ontology_cost_usd": 0.19278819860000002
    },
    {
      "domain": "ITBench",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 2,
      "tasks": [
        "Scenario-102",
        "Scenario-33"
      ],
      "reference_mean_score": 0.0,
      "ontology_mean_score": 0.0,
      "reference_latency_median_seconds": 88.97875841650239,
      "ontology_latency_median_seconds": 79.0887275210116,
      "reference_cost_usd": 0.1739302746,
      "ontology_cost_usd": 0.1090865338
    }
  ],
  "paired_pure_actor": [
    {
      "domain": "banking_knowledge",
      "profile": "reference",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "actor_mean_score": 0.0,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 820.1344427084987,
      "mithril_latency_median_seconds": 1147.1746780414978,
      "actor_cost_usd": 3.304979592,
      "mithril_cost_usd": 2.6200992602
    },
    {
      "domain": "banking_knowledge",
      "profile": "reference",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "actor_mean_score": 0.0,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 820.1344427084987,
      "mithril_latency_median_seconds": 1002.5419530624931,
      "actor_cost_usd": 3.304979592,
      "mithril_cost_usd": 3.3994619458
    },
    {
      "domain": "banking_knowledge",
      "profile": "ontology",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "actor_mean_score": 0.0,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 820.1344427084987,
      "mithril_latency_median_seconds": 918.3535884380108,
      "actor_cost_usd": 3.304979592,
      "mithril_cost_usd": 1.5048833684
    },
    {
      "domain": "banking_knowledge",
      "profile": "ontology",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "actor_mean_score": 0.0,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 820.1344427084987,
      "mithril_latency_median_seconds": 701.7196970415098,
      "actor_cost_usd": 3.304979592,
      "mithril_cost_usd": 1.1553809418
    },
    {
      "domain": "ITBench",
      "profile": "reference",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 3,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33"
      ],
      "actor_mean_score": 0.3333333333333333,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 82.525155041978,
      "mithril_latency_median_seconds": 204.2397644170269,
      "actor_cost_usd": 0.1140022608,
      "mithril_cost_usd": 0.3174566894
    },
    {
      "domain": "ITBench",
      "profile": "reference",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 3,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33"
      ],
      "actor_mean_score": 0.3333333333333333,
      "mithril_mean_score": 0.3333333333333333,
      "actor_latency_median_seconds": 82.525155041978,
      "mithril_latency_median_seconds": 89.79505529100425,
      "actor_cost_usd": 0.1140022608,
      "mithril_cost_usd": 0.271019634
    },
    {
      "domain": "ITBench",
      "profile": "ontology",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 3,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33"
      ],
      "actor_mean_score": 0.3333333333333333,
      "mithril_mean_score": 0.3333333333333333,
      "actor_latency_median_seconds": 82.525155041978,
      "mithril_latency_median_seconds": 80.14478270799736,
      "actor_cost_usd": 0.1140022608,
      "mithril_cost_usd": 0.1654750254
    },
    {
      "domain": "ITBench",
      "profile": "ontology",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 2,
      "tasks": [
        "Scenario-102",
        "Scenario-33"
      ],
      "actor_mean_score": 0.0,
      "mithril_mean_score": 0.0,
      "actor_latency_median_seconds": 81.73898870799167,
      "mithril_latency_median_seconds": 79.0887275210116,
      "actor_cost_usd": 0.0813496296,
      "mithril_cost_usd": 0.1090865338
    }
  ],
  "paired_cache": [
    {
      "domain": "banking_knowledge",
      "profile": "reference",
      "arm": "actor-only",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 3.304979592,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 820.1344427084987,
      "warm_latency_median_seconds": 42.121292291994905
    },
    {
      "domain": "banking_knowledge",
      "profile": "reference",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 2.6200992602,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 1147.1746780414978,
      "warm_latency_median_seconds": 39.983900916486164
    },
    {
      "domain": "banking_knowledge",
      "profile": "reference",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 3.3994619458,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 1002.5419530624931,
      "warm_latency_median_seconds": 59.591904083499685
    },
    {
      "domain": "banking_knowledge",
      "profile": "ontology",
      "arm": "actor-only",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 1.2087224382,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 454.96934816600697,
      "warm_latency_median_seconds": 35.683247187495
    },
    {
      "domain": "banking_knowledge",
      "profile": "ontology",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 1.5048833684,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 918.3535884380108,
      "warm_latency_median_seconds": 77.02185693749925
    },
    {
      "domain": "banking_knowledge",
      "profile": "ontology",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "task_033",
        "task_044",
        "task_065",
        "task_067"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 1.1553809418,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 701.7196970415098,
      "warm_latency_median_seconds": 62.23326622949389
    },
    {
      "domain": "ITBench",
      "profile": "reference",
      "arm": "actor-only",
      "complete_pairs": 3,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33"
      ],
      "same_score_pairs": 3,
      "primary_cost_usd": 0.1140022608,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 82.525155041978,
      "warm_latency_median_seconds": 1.8242214169877116
    },
    {
      "domain": "ITBench",
      "profile": "reference",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33",
        "Scenario-4"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 0.3727913106,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 150.26455737501965,
      "warm_latency_median_seconds": 4.687698625493795
    },
    {
      "domain": "ITBench",
      "profile": "reference",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 4,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33",
        "Scenario-4"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 0.346184481,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 89.4540095830016,
      "warm_latency_median_seconds": 1.8897731454926543
    },
    {
      "domain": "ITBench",
      "profile": "ontology",
      "arm": "actor-only",
      "complete_pairs": 4,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33",
        "Scenario-4"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 0.2126895804,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 112.01362068750313,
      "warm_latency_median_seconds": 1.7366541664960096
    },
    {
      "domain": "ITBench",
      "profile": "ontology",
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "complete_pairs": 4,
      "tasks": [
        "Scenario-102",
        "Scenario-12",
        "Scenario-33",
        "Scenario-4"
      ],
      "same_score_pairs": 4,
      "primary_cost_usd": 0.19278819860000002,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 88.35388874998898,
      "warm_latency_median_seconds": 4.034787354510627
    },
    {
      "domain": "ITBench",
      "profile": "ontology",
      "arm": "cloudflare/clef-flash",
      "complete_pairs": 2,
      "tasks": [
        "Scenario-102",
        "Scenario-33"
      ],
      "same_score_pairs": 2,
      "primary_cost_usd": 0.1090865338,
      "warm_cost_usd": 0,
      "primary_latency_median_seconds": 79.0887275210116,
      "warm_latency_median_seconds": 12.168910770997172
    }
  ],
  "provider_cache_verification": {
    "status": "partially-available",
    "verified_generation_receipts": 11,
    "read_only": true,
    "new_inference_calls": 0,
    "boundary": "Inference usage cached_tokens is the primary hit observation. Independently fetched generation receipts confirm identity, model/provider, invoice cost and any reported cache fields; omitted fields remain unknown. Earlier real GET receipts are retained with their original byte hashes when metadata expires, after matching physical study observations."
  },
  "accounting": {
    "study_stages": {
      "actor": {
        "confirmed_api_fees_usd": 4.980397459599997,
        "settled_responses": 473,
        "physical_api_seconds": 10785.414697912667,
        "dispatch_wait_seconds": 700.6544393300137,
        "backoff_seconds": 65
      },
      "user-simulator": {
        "confirmed_api_fees_usd": 0.12242870399999996,
        "settled_responses": 511,
        "physical_api_seconds": 4052.3152323142276,
        "dispatch_wait_seconds": 503.26365084786084,
        "backoff_seconds": 75
      },
      "review-and-revision": {
        "confirmed_api_fees_usd": 2.9431413758000016,
        "settled_responses": 584,
        "physical_api_seconds": 12437.246277829923,
        "dispatch_wait_seconds": 174.57526862982195,
        "backoff_seconds": 15
      },
      "working-ledger": {
        "confirmed_api_fees_usd": 0.0612528102,
        "settled_responses": 6,
        "physical_api_seconds": 887.4549496680556,
        "dispatch_wait_seconds": 0.04943074897164479,
        "backoff_seconds": 0
      },
      "actor-recovery": {
        "confirmed_api_fees_usd": 0.38809015620000004,
        "settled_responses": 32,
        "physical_api_seconds": 998.1006252939405,
        "dispatch_wait_seconds": 3.581348038977012,
        "backoff_seconds": 0
      },
      "concurrent-admission": {
        "confirmed_api_fees_usd": 0.000107445,
        "settled_responses": 3,
        "physical_api_seconds": 7.03193787598866,
        "dispatch_wait_seconds": 3.0150800829869695,
        "backoff_seconds": 0
      },
      "decision": {
        "confirmed_api_fees_usd": 0.061850011999999996,
        "settled_responses": 444,
        "physical_api_seconds": 508.1512448739668,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "it-actor": {
        "confirmed_api_fees_usd": 1.3373378897999997,
        "settled_responses": 136,
        "physical_api_seconds": 3133.121951210138,
        "dispatch_wait_seconds": 153.8699432888534,
        "backoff_seconds": 0
      },
      "it-revise-alternatives": {
        "confirmed_api_fees_usd": 0.3205833246,
        "settled_responses": 17,
        "physical_api_seconds": 511.1631005368545,
        "dispatch_wait_seconds": 0.12129817105596885,
        "backoff_seconds": 0
      },
      "it-critique": {
        "confirmed_api_fees_usd": 0.3484810092,
        "settled_responses": 22,
        "physical_api_seconds": 1089.0836470389622,
        "dispatch_wait_seconds": 0.19848404201911762,
        "backoff_seconds": 0
      },
      "native-context-canary": {
        "confirmed_api_fees_usd": 0.000239637,
        "settled_responses": 2,
        "physical_api_seconds": 2.0739830840029754,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "auto-tool": {
        "confirmed_api_fees_usd": 8.9175e-05,
        "settled_responses": 1,
        "physical_api_seconds": 2.278291458991589,
        "dispatch_wait_seconds": 8.666436083003646,
        "backoff_seconds": 0
      },
      "json-late-system": {
        "confirmed_api_fees_usd": 0.00020184,
        "settled_responses": 1,
        "physical_api_seconds": 6.319830957974773,
        "dispatch_wait_seconds": 0.002222750015789643,
        "backoff_seconds": 0
      },
      "review-and-revision-budget-admission": {
        "confirmed_api_fees_usd": 0.0082743372,
        "settled_responses": 2,
        "physical_api_seconds": 86.8957527919847,
        "dispatch_wait_seconds": 0.016183416009880602,
        "backoff_seconds": 0
      },
      "user-simulator-diagnostic": {
        "confirmed_api_fees_usd": 0.00082814,
        "settled_responses": 1,
        "physical_api_seconds": 4.503315250010928,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "native-routing-canary": {
        "confirmed_api_fees_usd": 5.715e-06,
        "settled_responses": 2,
        "physical_api_seconds": 1.4152139589714352,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "routing-canary": {
        "confirmed_api_fees_usd": 7.9e-05,
        "settled_responses": 1,
        "physical_api_seconds": 2.6574071669892874,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "source-control-decision": {
        "confirmed_api_fees_usd": 0.04745491799999996,
        "settled_responses": 341,
        "physical_api_seconds": 394.0736371220846,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0
      },
      "actor-joint-candidates": {
        "confirmed_api_fees_usd": 6.158446510200006,
        "settled_responses": 694,
        "physical_api_seconds": 8915.091086184897,
        "dispatch_wait_seconds": 92.01596238985076,
        "backoff_seconds": 30
      },
      "reasoning-disabled-exact-input-probe": {
        "confirmed_api_fees_usd": 0.0008646018,
        "settled_responses": 1,
        "physical_api_seconds": 26.18891770899063,
        "dispatch_wait_seconds": 0.0033156660210806876,
        "backoff_seconds": 0
      }
    },
    "study_confirmed_api_fees_usd": 16.7801540606,
    "study_uncertain_reserve_usd": 2.032207060000003,
    "shared_confirmed_api_fees_usd": 66.68010986560033,
    "shared_uncertain_reserve_usd": 2.133821560000003,
    "shared_cap_usd": 70,
    "new_settled_physical_responses": 3274,
    "boundary": "Invoice fees include provider caching; reserve is not treated as a zero-cost response. Hosting, local compute, labor, tax and foreign exchange excluded.",
    "primary_confirmed_api_fees_usd": 14.6755526688,
    "warm_confirmed_api_fees_usd": 0,
    "pilot_confirmed_api_fees_usd": 0.9467285046,
    "development_routing_and_other_confirmed_api_fees_usd": 1.157872887200001
  },
  "source_repair": {
    "at_utc": "2026-10-16T04:07:17.149999+00:00",
    "kind": "bounded-diagnostic-source-catalog",
    "observed_failure": "Original bank source indexing mistakenly submitted more than32diagnostic source-read entries as a single action catalog. All3ontology arms failed without an official reward.",
    "change": "Batch only implicit diagnostic source catalogs in groups of32. Keep the real action-catalog32limit, source facts/revision, candidate rules, text/native settings and all model instructions unchanged.",
    "existing_results": "Retain the three valid reference primary/warm results exactly. Reference code never invokes the changed ontology bridge. Never retry their evaluated zeros.",
    "continuation": "Fresh cold namespaces for the3unscored ontology conditions only, with all original failed primary/warm attempts, fees and attempted wall time retained in attempt_history. No cached model response imported from the failed namespace. Continue the same original preselected8task cohort; no outcome selection or new labels.",
    "qualification": "Original six known-case ontology controls (Actor, Decider, Clef in both domains), plus offline source-count tests at32,33,256 and unchanged real-action32guard. No win/score threshold."
  },
  "development_allocation_amendment": {
    "at_utc": "2026-10-16T04:16:53.666604+00:00",
    "old_internal_development_allocation_usd": 12,
    "new_internal_development_allocation_usd": 13,
    "total_human_approved_cap_usd": 60,
    "boundary": "Reallocate one dollar inside the existing total60cap to finish operational source-repair qualification. No increase to total authority, no fee/reserve reset, no evaluated selected-task reward retry. Development guard includes outstanding reservations, including earlier failed requests."
  },
  "budget_amendment": {
    "schema": "mithril.mimo-budget-amendment/v1",
    "prior_total_cap_usd": 60,
    "authorized_total_cap_usd": 65,
    "authorized_at_utc": "2026-10-16T05:54:54.985088+00:00",
    "human_instruction_sha256": "aedd392fca448f997fda7fbcc55b8fc5968920186576fe67f841400202276518",
    "source": "Direct human authorization; increase used only as required to complete the existing comparison.",
    "same_inference_source": true,
    "same_preselected_matrix": true,
    "same_models_and_settings": true,
    "prior_invoices_and_uncertain_reserves_preserved": true,
    "prior_confirmed_usd": 58.81791458520016,
    "prior_uncertain_reserve_usd": 0.9812594200000045,
    "boundary": "Only previously unscored financial interruptions and unattempted conditions continue. Evaluated rewards, including zeros, are preserved. No automatic increase above USD65."
  },
  "budget_amendment_latest": {
    "schema": "mithril.mimo-budget-amendment/v1",
    "prior_total_cap_usd": 65,
    "authorized_total_cap_usd": 70,
    "authorized_at_utc": "2026-10-16T07:26:59.506233+00:00",
    "human_instruction_sha256": "2647e6ad074f64e22a129ced1d380da0db98f1b3f80a0dc36a9db15e07b3d503",
    "prior_amendment_sha256": "296db9bac9d9987cfc2a69293216c8f18ea65971e9dc0aa0b0fcf9880b7e7fa9",
    "source": "Direct human authorization; increase used only as required to finish the existing comparison and unscored infrastructure recoveries.",
    "same_models_and_inputs": true,
    "same_preselected_matrix": true,
    "prior_invoices_and_uncertain_reserves_preserved": true,
    "prior_confirmed_usd": 63.42343290680023,
    "prior_uncertain_reserve_usd": 1.3391047000000031,
    "boundary": "No automatic increase above USD70. Evaluated rewards, including zeros, remain unchanged."
  },
  "admission_amendment": {
    "schema": "mithril.mimo-admission-amendment/v1",
    "recorded_at_utc": "2026-10-16T07:29:47.369637+00:00",
    "recorded_before_selected_IT_inference": true,
    "original_maximum_parallel_conditions": 3,
    "remaining_maximum_parallel_conditions": 1,
    "authorized_total_cap_usd": 70,
    "human_budget_amendment_sha256": "fe8a73d669065ac97b08b8dc25e5c6d8be7d1060e8e955e3749a4ae29e4a2df7",
    "reason": "Reduce simultaneous unsettled reservations under the unchanged human USD70 ceiling. All selected IT arms receive the same one-condition admission.",
    "same_original_matrix_and_arm_order": true,
    "same_inference_source": true,
    "same_models_inputs_seeds_tools_and_limits": true,
    "worker_lifecycle": "Fresh original per-condition Python worker for every primary; its immediate no-inference warm attempt shares that worker and source reader, as in the initial original matrix.",
    "recovery_order": "Within an already attempted bank block, financial interruptions continue before transient payload failures to preserve the large exact-response prefix. Unattempted IT order remains the original rotating order.",
    "financial_recovery": "Only explicit unscored financial interruptions or one proven provider 429/5xx payload transport error may continue in the same response namespace. Every prior fee and active-attempt wall is retained; valid rewards, including zeros, are never retried.",
    "timing_boundary": "Earlier bank blocks admitted up to three conditions. Remaining IT conditions admit one. These are observed end-to-end times under disclosed admission, not isolated model-speed or cross-domain estimates. No in-flight measurement was paused."
  },
  "method_summary": "The cumulative cap increased from USD60 to USD65, then conditionally to USD70, each only after direct human approval. A later Actor provider-payload 504 timeout remains unscored in its original receipt; one explicit infrastructure continuation preserves the original failed response and uncertain reserve outside the active successful-response cache. No zero fee is inferred from a failed generation lookup. Two initial USD60 financial interruptions and a later USD65 interruption continue only after their respective separate human authorizations, in their original exact-response namespaces; all earlier fees and active attempted wall times remain attributed. Valid rewards, including zeros, are never retried. Approval and coordination holds are excluded from active-attempt time. Three valid reference primary/warm pairs are preserved unchanged from the initial source epoch, including all evaluated zeros. Three unscored ontology source-index failures continue in fresh namespaces after batching only implicit diagnostic source catalogs; all original fees and attempted wall times are included. Actual action bounds and model instructions remain unchanged. Prospectively selected previously unused public tasks; matched MiMo/Novita text settings and original isolated environments. Bank blocks admit up to three conditions. Before any selected IT inference, a prospective admission amendment gives every IT arm one isolated condition at a time to reduce simultaneous uncertain reservations within the unchanged USD70 cap. The original matrix, rotating arm order, source inputs and model settings remain unchanged; each primary has a fresh original Python worker, with its immediate warm attempt sharing that worker. MiMo provider reasoning is explicitly disabled for every text role. Explicit critique/revision and native choices remain in the Mithril controller. Known-case low-effort requests exhausted 8,192 tokens with no visible output; a same-input non-reasoning probe returned usable source-checked JSON with 0 reported reasoning tokens. No unused-case inference preceded this common setting. The human-requested larger MiMo context uses a 262,144 cl100k proxy-token input guard,160,000 source-proxy budget with2× serialization allowance, up to 20,000-character original IT source excerpts and a larger768,000-character bank-document window. Same source and input bounds for every arm/profile; actual provider prompt_tokens are recorded. Native selection retains its separate common 8,192 proxy-token / 32,768-byte guard. The ontology treatment combines the observed-source graph, literal source index, historical immutable policy memory, stable prefix and native-first read/generate control. It is a joint controller treatment, not an isolated estimate of RDF typing. Consequential writes and final replies require MiMo. Actor + ontology is distinct from pure reference Actor. For IT, all arms use the same MiMo proposal schema with optional alternatives; Actor delivers the first diagnosis without critique or native selection. Native selectors run only for multiple distinct valid candidates. All attempts and physical fees remain recorded; missing/ambiguous grades remain null. Immediate warm replay has a 7200-second response TTL and prohibits new inference; its coverage and latency are not another independent quality trial. IT replay retains the immutable source index in the same worker; banking starts a fresh original simulation. Warm timing includes these lifecycle effects and is not an isolated estimate of application-cache lookup speed. The composed adapter uses the Mithril research loop and ActionOntology component, not every generic runSemanticStep path. The stopped earlier pilot remains separately disclosed, including two unscored outer timeouts and zero native requests. Its early banking runs experienced external compilation contention. New total condition times include actual provider queueing and original-environment initialization, with API waits and remaining wall time recorded separately. API-wait subtotals cover returned metered callbacks; residual wall can include unreturned or unsettled network waits and is not a CPU-only measure. These are observed end-to-end times, not an isolated model-speed benchmark. Older Gemini cohorts use different tasks and serial scheduling, so cross-cohort differences are not a pure model substitution effect.",
  "plan": {
    "schema": "mithril.mimo-source-scale-plan/v4",
    "created_at_utc": "2026-10-16T03:18:54.238465+00:00",
    "selection": "Retain the three preselected, unrun bank tasks from the stopped pilot and all four never-inferred IT incidents. Replace the used pilot task_078 with the next eligible ID by the original prospective SHA256 order; exclude all older used/planned cohorts. No outcome-based selection.",
    "bank_tasks": [
      "task_044",
      "task_033",
      "task_065",
      "task_067"
    ],
    "it_incidents": [
      "Scenario-4",
      "Scenario-12",
      "Scenario-33",
      "Scenario-102"
    ],
    "seeds": [
      613
    ],
    "arms": [
      "actor-only",
      "perplexity/pplx-decider-v1.1-27b",
      "cloudflare/clef-flash"
    ],
    "profiles": [
      "reference",
      "ontology"
    ],
    "primary_episodes": 48,
    "warm_episodes_maximum": 48,
    "bank_max_steps": 200,
    "actor_model": "xiaomi/mimo-v2.6-pro",
    "provider": "Novita",
    "provider_tag": "novita/fp8",
    "allow_fallbacks": false,
    "text_temperature": {
      "actor": 0.3,
      "critic": 0,
      "user-simulator": 0
    },
    "reasoning": {
      "enabled": false,
      "exclude": true
    },
    "development_only": [
      "task_034",
      "Scenario-11"
    ],
    "global_cap_usd": 60,
    "prior_study_confirmed_usd": 49.8999558050001,
    "prior_uncertain_reserve_usd": 0.1016145,
    "dataset_sha": "76df38a82288f75ba9e41dc8c515033332497473",
    "cache": {
      "primary": "Distinct fresh namespaces per task/seed/arm/profile; same TTL and no cross-arm/profile response reuse. Provider prompt cache is opportunistic and cannot be reset.",
      "warm": "Immediate same-namespace replay after each primary episode, strict no new physical inference. TTL expiration or changed requests remain explicit uncompleted warm attempts. Not an independent quality trial.",
      "ttl_seconds": 7200,
      "provider_cache": "Count actual reported metadata only; native unknown stays null. Invoiced fees include caching; no second discount.",
      "amendment": "New study only: bounded two-hour exact-response TTL, to cover a full original3600second episode plus its immediate replay. Applies to all arms/profiles; no old pilot cache reused. Full visible-state revision prevents partial-context collisions."
    },
    "scoring": "Original pinned tau2 per-task evaluator. Post-seal local deterministic IT root-group exact/filter/explicit-alias matching, not the official GPT-5.5 normalizer. Unsupported/ambiguous labels and execution failures remain null.",
    "latency": "Default reports total condition wall including initialization, original grading, dispatch/backoff and any allowed infrastructure continuation. Original episode and provider HTTP times remain separate.",
    "evaluation_blinding": "No new IT labels downloaded before every planned prediction is sealed. Bank private evaluator goals are provided only to the original user simulator/evaluator, not actor/ranker.",
    "profiles_specification": {
      "reference": "Actor-only uses the original native-tool text loop with shared source/progress/schema guards. Native arms use one common MiMo joint candidate call, consequential critique/revision, then native selection over genuinely distinct actions. No generated planning ledger.",
      "ontology": "Same original task, evidence access, validation and model limits. Adds the real ActionOntology observed-source types, literal source index, exact historical immutable policy memory, stable original prefix, and a native-first catalog of original read-only tool contracts or generate fallback. IT can select original source modalities. Account values are never restored as current state; writes/final replies require the common MiMo actor. Actor+ontology is an additional optimized control. This is a joint controller treatment, not an isolated causal estimate of RDF typing.",
      "boundary": "This measures the composed Python adapters using the Mithril research loop and ActionOntology. It does not validate every path of the generic experimental runSemanticStep controller or a global production coding-agent model switch."
    },
    "it_joint_proposal": {
      "all_arms_same_request": true,
      "optional_alternatives": 2,
      "description": "The same MiMo call may include source-checked distinct competing diagnoses. Actor delivers its primary best diagnosis; native arms can rank valid alternatives after critique. No padded candidates or gold labels."
    },
    "evaluator_metadata_audit": [
      {
        "task": "task_044",
        "reward_types": [
          "DB"
        ],
        "nl_assertions_count": 0
      },
      {
        "task": "task_033",
        "reward_types": [
          "ACTION"
        ],
        "nl_assertions_count": 0
      },
      {
        "task": "task_065",
        "reward_types": [
          "DB"
        ],
        "nl_assertions_count": 0
      },
      {
        "task": "task_067",
        "reward_types": [
          "DB"
        ],
        "nl_assertions_count": 0
      }
    ],
    "native_input_guard": {
      "whole_payload_cl100k_maximum": 8192,
      "whole_payload_bytes_maximum": 32768,
      "source_excerpts": "Budget available after complete choices/instructions; omitted sources explicitly unknown; no candidate truncation.",
      "boundary": "A common conservative proxy token/byte limit, not a claim to use the provider tokenizer."
    },
    "development_stage": {
      "known_cases": [
        "task_034",
        "Scenario-11"
      ],
      "conditions": [
        [
          "ontology",
          "actor-only"
        ],
        [
          "ontology",
          "perplexity/pplx-decider-v1.1-27b"
        ],
        [
          "ontology",
          "cloudflare/clef-flash"
        ]
      ],
      "max_steps": 24,
      "score_threshold": null,
      "qualification": "All6known original-environment ontology controls must complete; both native models must actually respond in each domain. Actor uses no native response. All text metadata must report0reasoning tokens and fit the256kproxy input guard. No win/score threshold."
    },
    "provider_amendment": {
      "at_utc": "2026-10-16T00:24:09.989452+00:00",
      "before_any_selected_task_inference": true,
      "from": "DeepInfra",
      "to": "Novita",
      "reason": "Captured upstream MiMo rate-limit responses on DeepInfra. Novita admits the same seeded FP8 MiMo model, reasoning, auto tools and JSON response format. One provider for every text role and all3arms; no fallback. A live operational probe is required before primary."
    },
    "text_dispatch_interval_seconds": 1,
    "text_completion_budget_tokens": 8192,
    "request_timeout_seconds": 180,
    "completion_budget_amendment": "Observed known-case MiMo critics exhausted2048tokens entirely in reasoning and returned no visible JSON. All text roles use a common8192total completion budget before any selected task; actual same-input critic probes must return usable JSON. Reasoning remains low/excluded; actual tokens/fees are metered.",
    "common_transport_admission": "Three concurrent unscored Novita JSON probes admitted with a common1second start floor and no HTTP error;429 cooldown/backoff unchanged. All original six15second-floor Novita known-case attempts retained; only operational failures recovered, with no quality threshold. Common transport admission, completion floor and request timeout changed before any unused-task inference.",
    "pilot_boundary": {
      "attempted": 3,
      "scored": 1,
      "native_requests": 0,
      "reason": "Both native arms reached outer timeout before any native call. Preserve pilot rewards, nulls, all fees/reserves separately."
    },
    "reasoning_amendment": {
      "before_any_unused_case_inference": true,
      "from": {
        "effort": "low",
        "exclude": true
      },
      "to": {
        "enabled": false,
        "exclude": true
      },
      "reason": "Known-case MiMo output consumed8192tokens in internal reasoning with empty content. Same-input/seed/model/Novita probe with reasoning disabled returned usable source-checked JSON and0reported reasoning tokens. Apply equally to all arms/profiles/text roles, then rerun live native/original-environment activation. All earlier probes/fees/reserves retained.",
      "quality_threshold_used": false
    },
    "text_context": {
      "input_proxy_limit_tokens": 262144,
      "model_provider_advertised_context_tokens": 1048576,
      "bank_document_character_budget": 768000,
      "shared_source_proxy_budget": 160000,
      "source_serialization_safety_factor": 2,
      "it_literal_record_maximum_characters": 20000,
      "initial_it_observations": "Expand the same preselected source IDs from original indexed text; no labels. Same input/source bounds for all3arms and bothprofiles.",
      "warm": "Reset the original immutable source-reader budget before each episode, including replay.",
      "boundary": "cl100k is a conservative common input-size proxy, not the MiMo tokenizer. The native whole-payload guard remains8192proxy tokens/32768bytes. Actual provider prompt_tokens are recorded."
    },
    "context_amendment": {
      "at_utc": "2026-10-16T03:23:33.577693+00:00",
      "direct_human_request": "mimo の context size をもっと増やして",
      "before_any_r6_inference": true,
      "source_policy": "Retain more original logs/objects/policies as literal data. No answer/evaluator input added. All comparison arms use the same larger text input and source budgets."
    },
    "initial_prospective_plan_sha256": "5461ca5c678eade075c486889ef3a67865c6ded7e0e736aa15e4e43ce028a406",
    "source_repair_amendment": {
      "at_utc": "2026-10-16T04:07:17.149999+00:00",
      "kind": "bounded-diagnostic-source-catalog",
      "observed_failure": "Original bank source indexing mistakenly submitted more than32diagnostic source-read entries as a single action catalog. All3ontology arms failed without an official reward.",
      "change": "Batch only implicit diagnostic source catalogs in groups of32. Keep the real action-catalog32limit, source facts/revision, candidate rules, text/native settings and all model instructions unchanged.",
      "existing_results": "Retain the three valid reference primary/warm results exactly. Reference code never invokes the changed ontology bridge. Never retry their evaluated zeros.",
      "continuation": "Fresh cold namespaces for the3unscored ontology conditions only, with all original failed primary/warm attempts, fees and attempted wall time retained in attempt_history. No cached model response imported from the failed namespace. Continue the same original preselected8task cohort; no outcome selection or new labels.",
      "qualification": "Original six known-case ontology controls (Actor, Decider, Clef in both domains), plus offline source-count tests at32,33,256 and unchanged real-action32guard. No win/score threshold."
    }
  },
  "pilot": {
    "status": "stopped-after-first-matched-block",
    "original_planned_conditions": 48,
    "attempted_conditions": 3,
    "scored_conditions": 1,
    "not_attempted_conditions": 45,
    "reason": "Neither configured selector received a native request in the real bank pilot; generated planning/critique dominated, and both exploration workers hit the outer3700second timeout without an original reward. A native-first source-catalog path must pass operational activation before more unused cases. Every result, fee and reserve is retained.",
    "it_labels_opened": false,
    "native_ranking_performance_established": false,
    "episodes": [
      {
        "domain": "banking_knowledge",
        "task": "task_078",
        "arm": "actor-only",
        "profile": "reference",
        "status": 0,
        "score": 1.0,
        "termination": "user_stop",
        "controller_wall_seconds": 1078.2036598749983,
        "metering": {
          "physical_text_responses": 51,
          "reported_completion_tokens": 23224,
          "reported_reasoning_tokens": 19267,
          "reasoning_metadata_responses": 51,
          "reasoning_metadata_exceeds_completion_responses": 0,
          "actual_cost_usd": 0.1921355058,
          "physical_responses": 51,
          "application_cache_hits": 0,
          "application_cache_denominator": 51,
          "provider_cache_hit_responses": 46,
          "provider_cache_metadata_responses": 51,
          "provider_cached_input_tokens": 416128,
          "provider_reported_prompt_tokens": 807927,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 962.8656430859701,
          "dispatch_wait_seconds": 2.3986613740271423,
          "backoff_seconds": 0,
          "observed_callback_seconds": 969.0743224991602
        },
        "native_requests_dispatched_or_prepared": 0
      },
      {
        "domain": "banking_knowledge",
        "task": "task_078",
        "arm": "perplexity/pplx-decider-v1.1-27b",
        "profile": "reference",
        "status": "outer-worker-timeout",
        "score": null,
        "termination": "outer-worker-timeout",
        "controller_wall_seconds": null,
        "configured_worker_timeout_seconds": 3700,
        "metering": {
          "physical_text_responses": 61,
          "reported_completion_tokens": 109053,
          "reported_reasoning_tokens": 102421,
          "reasoning_metadata_responses": 61,
          "reasoning_metadata_exceeds_completion_responses": 3,
          "actual_cost_usd": 0.3337441296,
          "physical_responses": 61,
          "application_cache_hits": 0,
          "application_cache_denominator": 61,
          "provider_cache_hit_responses": 53,
          "provider_cache_metadata_responses": 61,
          "provider_cached_input_tokens": 619136,
          "provider_reported_prompt_tokens": 1163134,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 3420.1700664237724,
          "dispatch_wait_seconds": 4.168317451083567,
          "backoff_seconds": 0,
          "observed_callback_seconds": 3430.0341322068125
        },
        "native_requests_dispatched_or_prepared": 0
      },
      {
        "domain": "banking_knowledge",
        "task": "task_078",
        "arm": "cloudflare/clef-flash",
        "profile": "reference",
        "status": "outer-worker-timeout",
        "score": null,
        "termination": "outer-worker-timeout",
        "controller_wall_seconds": null,
        "configured_worker_timeout_seconds": 3700,
        "metering": {
          "physical_text_responses": 57,
          "reported_completion_tokens": 104278,
          "reported_reasoning_tokens": 94475,
          "reasoning_metadata_responses": 57,
          "reasoning_metadata_exceeds_completion_responses": 1,
          "actual_cost_usd": 0.4208488692,
          "physical_responses": 57,
          "application_cache_hits": 0,
          "application_cache_denominator": 57,
          "provider_cache_hit_responses": 51,
          "provider_cache_metadata_responses": 57,
          "provider_cached_input_tokens": 329472,
          "provider_reported_prompt_tokens": 1085658,
          "native_decision_responses": 0,
          "provider_error_responses": 0,
          "physical_api_seconds": 3574.47311921019,
          "dispatch_wait_seconds": 1.7420726679556537,
          "backoff_seconds": 0,
          "observed_callback_seconds": 3581.319872253982
        },
        "native_requests_dispatched_or_prepared": 0
      }
    ]
  },
  "episodes": [
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 829.2067383750109,
      "episode_seconds": 791.6546388749848,
      "metering": {
        "maximum_text_input_proxy_tokens": 50670,
        "maximum_reported_text_prompt_tokens": 49996,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0031259195999999998,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.3476784726,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.34767847259999995,
        "physical_text_responses": 42,
        "reported_completion_tokens": 3266,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 42,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.3508043922,
        "physical_responses": 42,
        "application_cache_hits": 0,
        "application_cache_denominator": 42,
        "provider_cache_hit_responses": 37,
        "provider_cache_metadata_responses": 42,
        "provider_cached_input_tokens": 122752,
        "provider_reported_prompt_tokens": 921651,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 685.8409423770208,
        "dispatch_wait_seconds": 42.63010353798745,
        "backoff_seconds": 30,
        "observed_callback_seconds": 762.249774751981
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 7,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 28,
        "selected_tool_operations": 36,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 15,
          "get_current_time": 1,
          "get_credit_card_accounts_by_user": 1,
          "get_credit_card_transactions_by_user": 1,
          "grep": 11,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 3
        },
        "distinct_exact_tool_operations": 36,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 53.322107208980015,
      "episode_seconds": 17.469956208980875,
      "metering": {
        "maximum_text_input_proxy_tokens": 50670,
        "maximum_reported_text_prompt_tokens": 49996,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 42,
        "application_cache_denominator": 42,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.1461736679484602
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 7,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 28,
        "selected_tool_operations": 36,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 15,
          "get_current_time": 1,
          "get_credit_card_accounts_by_user": 1,
          "get_credit_card_transactions_by_user": 1,
          "grep": 11,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 3
        },
        "distinct_exact_tool_operations": 36,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 948.5004361669999,
      "episode_seconds": 908.7250392499845,
      "metering": {
        "maximum_text_input_proxy_tokens": 27937,
        "maximum_reported_text_prompt_tokens": 27009,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0031729812,
        "native_decision_cost_usd": 0.002335496,
        "actor_text_cost_usd": 0.251847609,
        "other_settled_cost_usd": 5.551115123125783e-17,
        "agent_side_cost_usd": 0.254183105,
        "physical_text_responses": 61,
        "reported_completion_tokens": 8360,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 61,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.2573560862,
        "physical_responses": 78,
        "application_cache_hits": 0,
        "application_cache_denominator": 78,
        "provider_cache_hit_responses": 56,
        "provider_cache_metadata_responses": 61,
        "provider_cached_input_tokens": 261632,
        "provider_reported_prompt_tokens": 829001,
        "native_decision_responses": 17,
        "provider_error_responses": 0,
        "physical_api_seconds": 794.4009485031129,
        "dispatch_wait_seconds": 43.63017507994664,
        "backoff_seconds": 30,
        "observed_callback_seconds": 874.6552912130719
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 26,
        "selected_tool_operations": 15,
        "operation_names": {
          "KB_search": 5,
          "get_user_information_by_name": 1,
          "get_credit_card_accounts_by_user": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 3
        },
        "distinct_exact_tool_operations": 15,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 38,
        "review_admitted_steps": 21,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 75.65840879199095,
      "episode_seconds": 35.74887504099752,
      "metering": {
        "maximum_text_input_proxy_tokens": 27937,
        "maximum_reported_text_prompt_tokens": 27009,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 78,
        "application_cache_denominator": 78,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.9663718341616914
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 26,
        "selected_tool_operations": 15,
        "operation_names": {
          "KB_search": 5,
          "get_user_information_by_name": 1,
          "get_credit_card_accounts_by_user": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 3
        },
        "distinct_exact_tool_operations": 15,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 38,
        "review_admitted_steps": 21,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 1166.9462610829796,
      "episode_seconds": 1131.0575532499934,
      "metering": {
        "maximum_text_input_proxy_tokens": 34980,
        "maximum_reported_text_prompt_tokens": 34066,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.004565685,
        "native_decision_cost_usd": 0.0026375,
        "actor_text_cost_usd": 0.3797512422,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.3823887422,
        "physical_text_responses": 69,
        "reported_completion_tokens": 10353,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 69,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.3869544272,
        "physical_responses": 89,
        "application_cache_hits": 0,
        "application_cache_denominator": 89,
        "provider_cache_hit_responses": 62,
        "provider_cache_metadata_responses": 69,
        "provider_cached_input_tokens": 343552,
        "provider_reported_prompt_tokens": 1203490,
        "native_decision_responses": 20,
        "provider_error_responses": 0,
        "physical_api_seconds": 1007.6659022007661,
        "dispatch_wait_seconds": 56.82267575309379,
        "backoff_seconds": 15,
        "observed_callback_seconds": 1086.3789217110316
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 29,
        "selected_tool_operations": 16,
        "operation_names": {
          "KB_search": 10,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "get_credit_card_accounts_by_user": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 1,
          "call_discoverable_agent_tool": 1
        },
        "distinct_exact_tool_operations": 16,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 52,
        "review_admitted_steps": 23,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 37.76590295799542,
      "episode_seconds": 17.744025624997448,
      "metering": {
        "maximum_text_input_proxy_tokens": 34980,
        "maximum_reported_text_prompt_tokens": 34066,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 89,
        "application_cache_denominator": 89,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.45104545491631143
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 29,
        "selected_tool_operations": 16,
        "operation_names": {
          "KB_search": 10,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "get_credit_card_accounts_by_user": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 1,
          "call_discoverable_agent_tool": 1
        },
        "distinct_exact_tool_operations": 16,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 52,
        "review_admitted_steps": 23,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 587.5983087079949,
      "episode_seconds": 414.4912643749849,
      "metering": {
        "maximum_text_input_proxy_tokens": 28199,
        "maximum_reported_text_prompt_tokens": 27235,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0015052169999999999,
        "native_decision_cost_usd": 0.003533458,
        "actor_text_cost_usd": 0.164521167,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.16805462499999999,
        "physical_text_responses": 64,
        "reported_completion_tokens": 9562,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 64,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.169559842,
        "physical_responses": 89,
        "application_cache_hits": 0,
        "application_cache_denominator": 89,
        "provider_cache_hit_responses": 64,
        "provider_cache_metadata_responses": 64,
        "provider_cached_input_tokens": 547840,
        "provider_reported_prompt_tokens": 905852,
        "native_decision_responses": 25,
        "provider_error_responses": 0,
        "physical_api_seconds": 481.5715337981237,
        "dispatch_wait_seconds": 10.21323970585945,
        "backoff_seconds": 0,
        "observed_callback_seconds": 496.6216364609718
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 25,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 25,
        "selected_tool_operations": 17,
        "operation_names": {
          "KB_search": 5,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 17,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 35,
        "review_admitted_steps": 16,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 74.01941216699197,
      "episode_seconds": 20.58366666600341,
      "metering": {
        "maximum_text_input_proxy_tokens": 28199,
        "maximum_reported_text_prompt_tokens": 27235,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 89,
        "application_cache_denominator": 89,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.43457491797744296
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 25,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 25,
        "selected_tool_operations": 17,
        "operation_names": {
          "KB_search": 5,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 17,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 35,
        "review_admitted_steps": 16,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 951.8654829590232,
      "episode_seconds": 712.4737909999967,
      "metering": {
        "maximum_text_input_proxy_tokens": 26174,
        "maximum_reported_text_prompt_tokens": 25079,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0015470765999999999,
        "native_decision_cost_usd": 0.00706754,
        "actor_text_cost_usd": 0.1801415454,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.1872090854,
        "physical_text_responses": 75,
        "reported_completion_tokens": 10892,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 75,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.188756162,
        "physical_responses": 129,
        "application_cache_hits": 0,
        "application_cache_denominator": 129,
        "provider_cache_hit_responses": 73,
        "provider_cache_metadata_responses": 75,
        "provider_cached_input_tokens": 632320,
        "provider_reported_prompt_tokens": 1022978,
        "native_decision_responses": 54,
        "provider_error_responses": 0,
        "physical_api_seconds": 840.0974054069084,
        "dispatch_wait_seconds": 6.725173376034945,
        "backoff_seconds": 0,
        "observed_callback_seconds": 852.4494297966303
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 24,
        "native_selected_reads": 1,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 24,
        "selected_tool_operations": 19,
        "operation_names": {
          "KB_search": 4,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "list_discoverable_agent_tools": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 5
        },
        "distinct_exact_tool_operations": 19,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 30,
        "review_admitted_steps": 19,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 110.36713004100602,
      "episode_seconds": 20.76790350000374,
      "metering": {
        "maximum_text_input_proxy_tokens": 26174,
        "maximum_reported_text_prompt_tokens": 25079,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 129,
        "application_cache_denominator": 129,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.42183467306313105
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 24,
        "native_selected_reads": 1,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 24,
        "selected_tool_operations": 19,
        "operation_names": {
          "KB_search": 4,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "list_discoverable_agent_tools": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 5
        },
        "distinct_exact_tool_operations": 19,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 30,
        "review_admitted_steps": 19,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 1188.3439631259826,
      "episode_seconds": 1082.0123822909954,
      "metering": {
        "maximum_text_input_proxy_tokens": 58729,
        "maximum_reported_text_prompt_tokens": 57812,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.002016987,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.5736897414,
        "other_settled_cost_usd": 1.1102230246251565e-16,
        "agent_side_cost_usd": 0.5736897414000001,
        "physical_text_responses": 74,
        "reported_completion_tokens": 5601,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 74,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.5757067284,
        "physical_responses": 74,
        "application_cache_hits": 0,
        "application_cache_denominator": 74,
        "provider_cache_hit_responses": 74,
        "provider_cache_metadata_responses": 74,
        "provider_cached_input_tokens": 953344,
        "provider_reported_prompt_tokens": 2257716,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 1073.0754152599839,
        "dispatch_wait_seconds": 6.252983252023114,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1083.2670951219916
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 9,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 41,
        "selected_tool_operations": 57,
        "operation_names": {
          "KB_search": 24,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "grep": 21,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 57,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 7,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_044",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 112.31812058397918,
      "episode_seconds": 25.195176957990043,
      "metering": {
        "maximum_text_input_proxy_tokens": 58729,
        "maximum_reported_text_prompt_tokens": 57812,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 74,
        "application_cache_denominator": 74,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.5317927048599813
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 10,
          "action_matches": 9,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 41,
        "selected_tool_operations": 57,
        "operation_names": {
          "KB_search": 24,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "grep": 21,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 57,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 7,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_repair": "Unscored source-index infrastructure failure retained; fresh namespace, unchanged model instructions and original task. Evaluated zeros never retried.",
      "source_epoch": "batched-diagnostic-source-catalog",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 105.24399670801358,
      "episode_seconds": 56.147386291006114,
      "metering": {
        "maximum_text_input_proxy_tokens": 12340,
        "maximum_reported_text_prompt_tokens": 12469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0008259504,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0071555496,
        "other_settled_cost_usd": 8.673617379884035e-19,
        "agent_side_cost_usd": 0.007155549600000001,
        "physical_text_responses": 12,
        "reported_completion_tokens": 816,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 12,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0079815,
        "physical_responses": 12,
        "application_cache_hits": 0,
        "application_cache_denominator": 12,
        "provider_cache_hit_responses": 12,
        "provider_cache_metadata_responses": 12,
        "provider_cached_input_tokens": 41600,
        "provider_reported_prompt_tokens": 57972,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 47.52129437302938,
        "dispatch_wait_seconds": 1.0635142500104848,
        "backoff_seconds": 0,
        "observed_callback_seconds": 49.78472525198595
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 7,
        "selected_tool_operations": 3,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 3,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 10,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 36.81401033399743,
      "episode_seconds": 2.1784507909906097,
      "metering": {
        "maximum_text_input_proxy_tokens": 12340,
        "maximum_reported_text_prompt_tokens": 12469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 12,
        "application_cache_denominator": 12,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.03266466502100229
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 7,
        "selected_tool_operations": 3,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 3,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 10,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 214.36618566699326,
      "episode_seconds": 179.20200258298428,
      "metering": {
        "maximum_text_input_proxy_tokens": 14412,
        "maximum_reported_text_prompt_tokens": 13850,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0012318792,
        "native_decision_cost_usd": 0.00137982,
        "actor_text_cost_usd": 0.030734811600000002,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0321146316,
        "physical_text_responses": 16,
        "reported_completion_tokens": 1728,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 16,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0333465108,
        "physical_responses": 27,
        "application_cache_hits": 0,
        "application_cache_denominator": 27,
        "provider_cache_hit_responses": 14,
        "provider_cache_metadata_responses": 16,
        "provider_cached_input_tokens": 76928,
        "provider_reported_prompt_tokens": 146322,
        "native_decision_responses": 11,
        "provider_error_responses": 0,
        "physical_api_seconds": 157.76396929408656,
        "dispatch_wait_seconds": 2.2682005419919733,
        "backoff_seconds": 0,
        "observed_callback_seconds": 161.37220849696314
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 8,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 8,
        "selected_tool_operations": 4,
        "operation_names": {
          "KB_search": 2,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 4,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 20,
        "review_admitted_steps": 5,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 35.782987916027196,
      "episode_seconds": 10.973135083011584,
      "metering": {
        "maximum_text_input_proxy_tokens": 14412,
        "maximum_reported_text_prompt_tokens": 13850,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 27,
        "application_cache_denominator": 27,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.13086195598589256
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 8,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 8,
        "selected_tool_operations": 4,
        "operation_names": {
          "KB_search": 2,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 4,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 20,
        "review_admitted_steps": 5,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 274.25987183398684,
      "episode_seconds": 244.07072704200982,
      "metering": {
        "maximum_text_input_proxy_tokens": 20843,
        "maximum_reported_text_prompt_tokens": 20152,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.001549167,
        "native_decision_cost_usd": 0.00349198,
        "actor_text_cost_usd": 0.0614590452,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0649510252,
        "physical_text_responses": 33,
        "reported_completion_tokens": 3822,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 33,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0665001922,
        "physical_responses": 56,
        "application_cache_hits": 0,
        "application_cache_denominator": 56,
        "provider_cache_hit_responses": 33,
        "provider_cache_metadata_responses": 33,
        "provider_cached_input_tokens": 292352,
        "provider_reported_prompt_tokens": 427135,
        "native_decision_responses": 23,
        "provider_error_responses": 0,
        "physical_api_seconds": 207.15037208324065,
        "dispatch_wait_seconds": 3.769909752852982,
        "backoff_seconds": 0,
        "observed_callback_seconds": 213.0358423300204
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 14,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 5,
        "operation_names": {
          "KB_search": 3,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 5,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 29,
        "review_admitted_steps": 11,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 22.300239041011082,
      "episode_seconds": 9.9355060000089,
      "metering": {
        "maximum_text_input_proxy_tokens": 20843,
        "maximum_reported_text_prompt_tokens": 20152,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 56,
        "application_cache_denominator": 56,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.14570578795974143
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 14,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 5,
        "operation_names": {
          "KB_search": 3,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 5,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 29,
        "review_admitted_steps": 11,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 133.2974685410154,
      "episode_seconds": 115.01162000000477,
      "metering": {
        "maximum_text_input_proxy_tokens": 12074,
        "maximum_reported_text_prompt_tokens": 12206,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0011102964,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0525348114,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0525348114,
        "physical_text_responses": 24,
        "reported_completion_tokens": 1184,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 24,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0536451078,
        "physical_responses": 24,
        "application_cache_hits": 0,
        "application_cache_denominator": 24,
        "provider_cache_hit_responses": 24,
        "provider_cache_metadata_responses": 24,
        "provider_cached_input_tokens": 69248,
        "provider_reported_prompt_tokens": 189629,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 98.99218870489858,
        "dispatch_wait_seconds": 11.336776250042021,
        "backoff_seconds": 0,
        "observed_callback_seconds": 110.96507816604571
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 13,
        "selected_tool_operations": 7,
        "operation_names": {
          "KB_search": 1,
          "grep": 1,
          "get_current_time": 1,
          "unlock_discoverable_agent_tool": 2,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 7,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 17,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 18.564240999985486,
      "episode_seconds": 4.186822584015317,
      "metering": {
        "maximum_text_input_proxy_tokens": 12074,
        "maximum_reported_text_prompt_tokens": 12206,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 24,
        "application_cache_denominator": 24,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.12219733101665042
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 13,
        "selected_tool_operations": 7,
        "operation_names": {
          "KB_search": 1,
          "grep": 1,
          "get_current_time": 1,
          "unlock_discoverable_agent_tool": 2,
          "get_user_information_by_name": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 7,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 17,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 322.27156908399775,
      "episode_seconds": 303.5643621250056,
      "metering": {
        "maximum_text_input_proxy_tokens": 13037,
        "maximum_reported_text_prompt_tokens": 12330,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.000941601,
        "native_decision_cost_usd": 0.000520129,
        "actor_text_cost_usd": 0.0676875342,
        "other_settled_cost_usd": 1.3877787807814457e-17,
        "agent_side_cost_usd": 0.0682076632,
        "physical_text_responses": 39,
        "reported_completion_tokens": 4659,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 39,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0691492642,
        "physical_responses": 45,
        "application_cache_hits": 0,
        "application_cache_denominator": 45,
        "provider_cache_hit_responses": 37,
        "provider_cache_metadata_responses": 39,
        "provider_cached_input_tokens": 95232,
        "provider_reported_prompt_tokens": 242894,
        "native_decision_responses": 6,
        "provider_error_responses": 0,
        "physical_api_seconds": 289.3752464988793,
        "dispatch_wait_seconds": 7.358651457063388,
        "backoff_seconds": 0,
        "observed_callback_seconds": 297.64584354302497
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 5,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 1,
          "unlock_discoverable_agent_tool": 2,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 5,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 10,
        "review_admitted_steps": 13,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 17.23074358300073,
      "episode_seconds": 4.017044625012204,
      "metering": {
        "maximum_text_input_proxy_tokens": 13037,
        "maximum_reported_text_prompt_tokens": 12330,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 45,
        "application_cache_denominator": 45,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.13952500201412477
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 5,
        "operation_names": {
          "get_user_information_by_name": 1,
          "KB_search": 1,
          "unlock_discoverable_agent_tool": 2,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 5,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 10,
        "review_admitted_steps": 13,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 376.7601704580011,
      "episode_seconds": 358.78942366698175,
      "metering": {
        "maximum_text_input_proxy_tokens": 19910,
        "maximum_reported_text_prompt_tokens": 19163,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0009934397999999999,
        "native_decision_cost_usd": 0.00126004,
        "actor_text_cost_usd": 0.0929077044,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0941677444,
        "physical_text_responses": 36,
        "reported_completion_tokens": 4389,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 36,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0951611842,
        "physical_responses": 46,
        "application_cache_hits": 0,
        "application_cache_denominator": 46,
        "provider_cache_hit_responses": 35,
        "provider_cache_metadata_responses": 36,
        "provider_cached_input_tokens": 137472,
        "provider_reported_prompt_tokens": 343421,
        "native_decision_responses": 10,
        "provider_error_responses": 0,
        "physical_api_seconds": 346.6368234149995,
        "dispatch_wait_seconds": 5.214411044958979,
        "backoff_seconds": 0,
        "observed_callback_seconds": 352.9099737529468
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 6,
        "operation_names": {
          "KB_search": 3,
          "get_user_information_by_name": 1,
          "get_user_information_by_email": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 6,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 25,
        "review_admitted_steps": 13,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_033",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 21.140435290988535,
      "episode_seconds": 4.906864999997197,
      "metering": {
        "maximum_text_input_proxy_tokens": 19910,
        "maximum_reported_text_prompt_tokens": 19163,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 46,
        "application_cache_denominator": 46,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.13268791395239532
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 5,
          "action_matches": 1,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 14,
        "selected_tool_operations": 6,
        "operation_names": {
          "KB_search": 3,
          "get_user_information_by_name": 1,
          "get_user_information_by_email": 1,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 6,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 0,
        "last_distinct_source_fragments": 25,
        "review_admitted_steps": 13,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 811.0621470419865,
      "episode_seconds": 792.2702125000069,
      "metering": {
        "maximum_text_input_proxy_tokens": 64019,
        "maximum_reported_text_prompt_tokens": 63334,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.002728032,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.9388003452,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.9388003452,
        "physical_text_responses": 57,
        "reported_completion_tokens": 6411,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 57,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.9415283772,
        "physical_responses": 57,
        "application_cache_hits": 0,
        "application_cache_denominator": 57,
        "provider_cache_hit_responses": 55,
        "provider_cache_metadata_responses": 57,
        "provider_cached_input_tokens": 183552,
        "provider_reported_prompt_tokens": 2333644,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 764.0483695410076,
        "dispatch_wait_seconds": 7.194750958966324,
        "backoff_seconds": 0,
        "observed_callback_seconds": 772.8941218310501
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 38,
        "selected_tool_operations": 81,
        "operation_names": {
          "KB_search": 51,
          "grep": 16,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 81,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 6,
        "last_distinct_source_fragments": 96,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 30.920477375009796,
      "episode_seconds": 17.36855208297493,
      "metering": {
        "maximum_text_input_proxy_tokens": 64019,
        "maximum_reported_text_prompt_tokens": 63334,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 57,
        "application_cache_denominator": 57,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.1305911710660439
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 38,
        "selected_tool_operations": 81,
        "operation_names": {
          "KB_search": 51,
          "grep": 16,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 81,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 6,
        "last_distinct_source_fragments": 96,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 1056.5834699579864,
      "episode_seconds": 1036.8742009999987,
      "metering": {
        "maximum_text_input_proxy_tokens": 41982,
        "maximum_reported_text_prompt_tokens": 40842,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0027008694,
        "native_decision_cost_usd": 0.004755347,
        "actor_text_cost_usd": 0.7360910952,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.7408464422,
        "physical_text_responses": 98,
        "reported_completion_tokens": 18601,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 98,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.7435473116,
        "physical_responses": 129,
        "application_cache_hits": 0,
        "application_cache_denominator": 129,
        "provider_cache_hit_responses": 97,
        "provider_cache_metadata_responses": 98,
        "provider_cached_input_tokens": 628736,
        "provider_reported_prompt_tokens": 2284703,
        "native_decision_responses": 31,
        "provider_error_responses": 0,
        "physical_api_seconds": 1004.3833090785483,
        "dispatch_wait_seconds": 6.55416433836217,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1013.9477541320084
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 43,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 16,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 28,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 55,
        "review_admitted_steps": 36,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 43.52539937500842,
      "episode_seconds": 23.269512375001796,
      "metering": {
        "maximum_text_input_proxy_tokens": 41982,
        "maximum_reported_text_prompt_tokens": 40842,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 129,
        "application_cache_denominator": 129,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.6204748440068215
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 43,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 16,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 28,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 55,
        "review_admitted_steps": 36,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 1127.403095000016,
      "episode_seconds": 1108.704796624981,
      "metering": {
        "maximum_text_input_proxy_tokens": 44340,
        "maximum_reported_text_prompt_tokens": 43058,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.003035904,
        "native_decision_cost_usd": 0.00422224,
        "actor_text_cost_usd": 0.8634812574,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.8677034974000001,
        "physical_text_responses": 106,
        "reported_completion_tokens": 18204,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 106,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.8707394014000001,
        "physical_responses": 138,
        "application_cache_hits": 0,
        "application_cache_denominator": 138,
        "provider_cache_hit_responses": 103,
        "provider_cache_metadata_responses": 106,
        "provider_cached_input_tokens": 743424,
        "provider_reported_prompt_tokens": 2692857,
        "native_decision_responses": 32,
        "provider_error_responses": 0,
        "physical_api_seconds": 1070.8591364632011,
        "dispatch_wait_seconds": 6.645334542845376,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1081.5521881298628
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 48,
        "selected_tool_operations": 32,
        "operation_names": {
          "KB_search": 17,
          "grep": 1,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 58,
        "review_admitted_steps": 39,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 42.201898874976905,
      "episode_seconds": 24.062548500020057,
      "metering": {
        "maximum_text_input_proxy_tokens": 44340,
        "maximum_reported_text_prompt_tokens": 43058,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 138,
        "application_cache_denominator": 138,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.5462491681100801
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 48,
        "selected_tool_operations": 32,
        "operation_names": {
          "KB_search": 17,
          "grep": 1,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 58,
        "review_admitted_steps": 39,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 379.15071629101294,
      "episode_seconds": 362.8114104580018,
      "metering": {
        "maximum_text_input_proxy_tokens": 49282,
        "maximum_reported_text_prompt_tokens": 49087,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0014111394,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.413458194,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.413458194,
        "physical_text_responses": 59,
        "reported_completion_tokens": 3597,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 59,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.4148693334,
        "physical_responses": 59,
        "application_cache_hits": 0,
        "application_cache_denominator": 59,
        "provider_cache_hit_responses": 58,
        "provider_cache_metadata_responses": 59,
        "provider_cached_input_tokens": 793344,
        "provider_reported_prompt_tokens": 1733307,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 339.49761495602434,
        "dispatch_wait_seconds": 4.633671124931425,
        "backoff_seconds": 0,
        "observed_callback_seconds": 345.63546341494657
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 49,
        "selected_tool_operations": 42,
        "operation_names": {
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "KB_search": 26,
          "grep": 2,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 40,
        "repeated_exact_tool_operations": 2,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 85,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 34.55248404099257,
      "episode_seconds": 15.595666542008985,
      "metering": {
        "maximum_text_input_proxy_tokens": 49282,
        "maximum_reported_text_prompt_tokens": 49087,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 59,
        "application_cache_denominator": 59,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.2048028689750936
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 49,
        "selected_tool_operations": 42,
        "operation_names": {
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "KB_search": 26,
          "grep": 2,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 40,
        "repeated_exact_tool_operations": 2,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 85,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 815.8410853750247,
      "episode_seconds": 797.496943584003,
      "metering": {
        "maximum_text_input_proxy_tokens": 42208,
        "maximum_reported_text_prompt_tokens": 40806,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.003392844,
        "native_decision_cost_usd": 0.005926317,
        "actor_text_cost_usd": 0.406100583,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.4120269,
        "physical_text_responses": 92,
        "reported_completion_tokens": 16155,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 92,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.415419744,
        "physical_responses": 133,
        "application_cache_hits": 0,
        "application_cache_denominator": 133,
        "provider_cache_hit_responses": 87,
        "provider_cache_metadata_responses": 92,
        "provider_cached_input_tokens": 1521920,
        "provider_reported_prompt_tokens": 2418379,
        "native_decision_responses": 41,
        "provider_error_responses": 0,
        "physical_api_seconds": 752.8486227451358,
        "dispatch_wait_seconds": 10.658391501987353,
        "backoff_seconds": 0,
        "observed_callback_seconds": 766.3807892850309
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 43,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 43,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 11,
          "grep": 2,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 27,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 52,
        "review_admitted_steps": 30,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 50.447120291995816,
      "episode_seconds": 32.9336994169862,
      "metering": {
        "maximum_text_input_proxy_tokens": 42208,
        "maximum_reported_text_prompt_tokens": 40806,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 133,
        "application_cache_denominator": 133,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.457804245990701
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 43,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 43,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 11,
          "grep": 2,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 27,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 52,
        "review_admitted_steps": 30,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 884.8416939169983,
      "episode_seconds": 867.8744141249917,
      "metering": {
        "maximum_text_input_proxy_tokens": 46213,
        "maximum_reported_text_prompt_tokens": 44914,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0021636299999999997,
        "native_decision_cost_usd": 0.00914878,
        "actor_text_cost_usd": 0.456426069,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.465574849,
        "physical_text_responses": 87,
        "reported_completion_tokens": 17519,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 87,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.467738479,
        "physical_responses": 155,
        "application_cache_hits": 0,
        "application_cache_denominator": 155,
        "provider_cache_hit_responses": 84,
        "provider_cache_metadata_responses": 87,
        "provider_cached_input_tokens": 1411840,
        "provider_reported_prompt_tokens": 2419347,
        "native_decision_responses": 68,
        "provider_error_responses": 0,
        "physical_api_seconds": 818.2959143310436,
        "dispatch_wait_seconds": 9.603801546036266,
        "backoff_seconds": 0,
        "observed_callback_seconds": 831.1112708311994
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 41,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 41,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 14,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 6,
          "grep": 1
        },
        "distinct_exact_tool_operations": 27,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 68,
        "review_admitted_steps": 33,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_065",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 52.04364916699706,
      "episode_seconds": 36.53638995799702,
      "metering": {
        "maximum_text_input_proxy_tokens": 46213,
        "maximum_reported_text_prompt_tokens": 44914,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 155,
        "application_cache_denominator": 155,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.4052219220029656
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 6,
          "action_matches": 3,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 41,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 41,
        "selected_tool_operations": 28,
        "operation_names": {
          "KB_search": 14,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 6,
          "grep": 1
        },
        "distinct_exact_tool_operations": 27,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 68,
        "review_admitted_steps": 33,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 2792.038303832989,
      "episode_seconds": 2353.4252128340013,
      "metering": {
        "maximum_text_input_proxy_tokens": 55493,
        "maximum_reported_text_prompt_tokens": 54055,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0031799652,
        "native_decision_cost_usd": 0.01298048,
        "actor_text_cost_usd": 0.7988817714,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.8118622514,
        "physical_text_responses": 142,
        "reported_completion_tokens": 25950,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 142,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.8150422166,
        "physical_responses": 239,
        "application_cache_hits": 65,
        "application_cache_denominator": 304,
        "provider_cache_hit_responses": 139,
        "provider_cache_metadata_responses": 142,
        "provider_cached_input_tokens": 2697856,
        "provider_reported_prompt_tokens": 4467449,
        "native_decision_responses": 97,
        "provider_error_responses": 0,
        "physical_api_seconds": 2647.362700033089,
        "dispatch_wait_seconds": 6.460920540703228,
        "backoff_seconds": 0,
        "observed_callback_seconds": 2660.6853493712842
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 58,
        "native_selected_reads": 1,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 58,
        "selected_tool_operations": 39,
        "operation_names": {
          "KB_search": 22,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "grep": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 37,
        "repeated_exact_tool_operations": 2,
        "maximum_no_new_content_streak": 3,
        "last_distinct_source_fragments": 72,
        "review_admitted_steps": 46,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 102.00006470800145,
      "episode_seconds": 58.85596954100765,
      "metering": {
        "maximum_text_input_proxy_tokens": 55493,
        "maximum_reported_text_prompt_tokens": 54055,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 304,
        "application_cache_denominator": 304,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1.162439675128553
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 58,
        "native_selected_reads": 1,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 58,
        "selected_tool_operations": 39,
        "operation_names": {
          "KB_search": 22,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "grep": 1,
          "get_credit_card_accounts_by_user": 1,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 7
        },
        "distinct_exact_tool_operations": 37,
        "repeated_exact_tool_operations": 2,
        "maximum_no_new_content_streak": 3,
        "last_distinct_source_fragments": 72,
        "review_admitted_steps": 46,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 530.787980041001,
      "episode_seconds": 506.186121457984,
      "metering": {
        "maximum_text_input_proxy_tokens": 52577,
        "maximum_reported_text_prompt_tokens": 52242,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.004296747,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.2058681294,
        "other_settled_cost_usd": 2.7755575615628914e-17,
        "agent_side_cost_usd": 0.20586812940000002,
        "physical_text_responses": 51,
        "reported_completion_tokens": 6811,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 51,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.2101648764,
        "physical_responses": 51,
        "application_cache_hits": 0,
        "application_cache_denominator": 51,
        "provider_cache_hit_responses": 50,
        "provider_cache_metadata_responses": 51,
        "provider_cached_input_tokens": 989824,
        "provider_reported_prompt_tokens": 1451148,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 482.47575783295906,
        "dispatch_wait_seconds": 2.948047871061135,
        "backoff_seconds": 0,
        "observed_callback_seconds": 487.0702885009814
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 36,
        "selected_tool_operations": 33,
        "operation_names": {
          "KB_search": 17,
          "grep": 3,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 89,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 26.737829374993453,
      "episode_seconds": 11.691036625008564,
      "metering": {
        "maximum_text_input_proxy_tokens": 52577,
        "maximum_reported_text_prompt_tokens": 52242,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 51,
        "application_cache_denominator": 51,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.16255745993112214
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 36,
        "selected_tool_operations": 33,
        "operation_names": {
          "KB_search": 17,
          "grep": 3,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 6
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 89,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 1718.546956457023,
      "episode_seconds": 494.0874032499851,
      "metering": {
        "maximum_text_input_proxy_tokens": 53895,
        "maximum_reported_text_prompt_tokens": 52417,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0035209878,
        "native_decision_cost_usd": 0.009834178,
        "actor_text_cost_usd": 0.4905459978,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.5003801758,
        "physical_text_responses": 103,
        "reported_completion_tokens": 19726,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 103,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.5039011636,
        "physical_responses": 170,
        "application_cache_hits": 119,
        "application_cache_denominator": 289,
        "provider_cache_hit_responses": 102,
        "provider_cache_metadata_responses": 103,
        "provider_cached_input_tokens": 2180096,
        "provider_reported_prompt_tokens": 3258388,
        "native_decision_responses": 67,
        "provider_error_responses": 0,
        "physical_api_seconds": 1570.5077050348627,
        "dispatch_wait_seconds": 3.11844492299133,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1579.2020797942532
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 50,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 50,
        "selected_tool_operations": 36,
        "operation_names": {
          "KB_search": 14,
          "get_user_information_by_name": 1,
          "grep": 4,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 10,
          "get_current_time": 1,
          "log_verification": 1
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 4,
        "maximum_no_new_content_streak": 4,
        "last_distinct_source_fragments": 72,
        "review_admitted_steps": 32,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 113.42678079201141,
      "episode_seconds": 40.56902412499767,
      "metering": {
        "maximum_text_input_proxy_tokens": 53895,
        "maximum_reported_text_prompt_tokens": 52417,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 289,
        "application_cache_denominator": 289,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.9368399003869854
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 4,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 50,
        "native_selected_reads": 2,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 50,
        "selected_tool_operations": 36,
        "operation_names": {
          "KB_search": 14,
          "get_user_information_by_name": 1,
          "grep": 4,
          "unlock_discoverable_agent_tool": 5,
          "call_discoverable_agent_tool": 10,
          "get_current_time": 1,
          "log_verification": 1
        },
        "distinct_exact_tool_operations": 32,
        "repeated_exact_tool_operations": 4,
        "maximum_no_new_content_streak": 4,
        "last_distinct_source_fragments": 72,
        "review_admitted_steps": 32,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 4734.367748749995,
      "episode_seconds": 2237.499965416995,
      "metering": {
        "maximum_text_input_proxy_tokens": 105191,
        "maximum_reported_text_prompt_tokens": 104731,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0055997604,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 1.9534019544,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 1.9534019544,
        "physical_text_responses": 71,
        "reported_completion_tokens": 11291,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 71,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 1.9590017148,
        "physical_responses": 71,
        "application_cache_hits": 29,
        "application_cache_denominator": 100,
        "provider_cache_hit_responses": 60,
        "provider_cache_metadata_responses": 71,
        "provider_cached_input_tokens": 211968,
        "provider_reported_prompt_tokens": 4691084,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 4090.228555326001,
        "dispatch_wait_seconds": 6.115052669978468,
        "backoff_seconds": 30,
        "observed_callback_seconds": 4138.855324252829
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 45,
        "selected_tool_operations": 144,
        "operation_names": {
          "KB_search": 100,
          "grep": 34,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 144,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 11,
        "last_distinct_source_fragments": 164,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "One unscored infrastructure continuation for an explicit provider 429/5xx error embedded in an HTTP200 body. Same task, model, input settings, seed and response namespace. All earlier fees, uncertain reserves and active-attempt wall times remain included. Evaluated rewards are never retried.",
      "transient_payload_recovery": true,
      "transient_payload_proof": [
        {
          "request_sha256": "1d4712f3ad2aacb9cee8cffe5da1df51489f5182fbd8feeacfaca00e713110d2",
          "raw_response_sha256": "cd5492d6c1cf18f51dd26df0087fadb706f0b7a8b0d6314a68af080ca5330c00",
          "payload_error_code": 504,
          "unsettled_reserve_usd": 0.35784528,
          "boundary": "Original response/timing preserved outside the active successful-response cache. The original uncertain reserve remains unchanged. One fresh infrastructure continuation is allowed; no invoice or zero fee is invented."
        }
      ],
      "source_epoch": "unchanged-inference-source-transient-payload-recovery",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 269.3310895420145,
      "episode_seconds": 52.89351504098158,
      "metering": {
        "maximum_text_input_proxy_tokens": 105191,
        "maximum_reported_text_prompt_tokens": 104731,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 100,
        "application_cache_denominator": 100,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 1.1010792457964271
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 45,
        "selected_tool_operations": 144,
        "operation_names": {
          "KB_search": 100,
          "grep": 34,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 3,
          "call_discoverable_agent_tool": 4
        },
        "distinct_exact_tool_operations": 144,
        "repeated_exact_tool_operations": 0,
        "maximum_no_new_content_streak": 11,
        "last_distinct_source_fragments": 164,
        "review_admitted_steps": 0,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "controller_recovery": "One unscored infrastructure continuation for an explicit provider 429/5xx error embedded in an HTTP200 body. Same task, model, input settings, seed and response namespace. All earlier fees, uncertain reserves and active-attempt wall times remain included. Evaluated rewards are never retried.",
      "transient_payload_recovery": true,
      "transient_payload_proof": [
        {
          "request_sha256": "1d4712f3ad2aacb9cee8cffe5da1df51489f5182fbd8feeacfaca00e713110d2",
          "raw_response_sha256": "cd5492d6c1cf18f51dd26df0087fadb706f0b7a8b0d6314a68af080ca5330c00",
          "payload_error_code": 504,
          "unsettled_reserve_usd": 0.35784528,
          "boundary": "Original response/timing preserved outside the active successful-response cache. The original uncertain reserve remains unchanged. One fresh infrastructure continuation is allowed; no invoice or zero fee is invented."
        }
      ],
      "source_epoch": "unchanged-inference-source-transient-payload-recovery",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 2179.8940033339895,
      "episode_seconds": 2157.9204345420003,
      "metering": {
        "maximum_text_input_proxy_tokens": 50785,
        "maximum_reported_text_prompt_tokens": 49690,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.0061703742,
        "native_decision_cost_usd": 0.00645298,
        "actor_text_cost_usd": 1.2546208932,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 1.2610738732,
        "physical_text_responses": 156,
        "reported_completion_tokens": 37992,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 156,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 1.2672442474,
        "physical_responses": 204,
        "application_cache_hits": 0,
        "application_cache_denominator": 204,
        "provider_cache_hit_responses": 151,
        "provider_cache_metadata_responses": 156,
        "provider_cached_input_tokens": 1382784,
        "provider_reported_prompt_tokens": 4193727,
        "native_decision_responses": 48,
        "provider_error_responses": 0,
        "physical_api_seconds": 2094.3048797460215,
        "dispatch_wait_seconds": 4.73689775398816,
        "backoff_seconds": 0,
        "observed_callback_seconds": 2105.294874585845
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 64,
        "selected_tool_operations": 36,
        "operation_names": {
          "KB_search": 21,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 7,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 35,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 64,
        "review_admitted_steps": 55,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "user_stop",
      "controller_wall_seconds": 49.42008995800279,
      "episode_seconds": 32.79379766600323,
      "metering": {
        "maximum_text_input_proxy_tokens": 50785,
        "maximum_reported_text_prompt_tokens": 49690,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 204,
        "application_cache_denominator": 204,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.593059047969291
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": false,
          "action_checks": 9,
          "action_matches": 5,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 64,
        "selected_tool_operations": 36,
        "operation_names": {
          "KB_search": 21,
          "get_user_information_by_name": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 4,
          "call_discoverable_agent_tool": 7,
          "transfer_to_human_agents": 1
        },
        "distinct_exact_tool_operations": 35,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 1,
        "last_distinct_source_fragments": 64,
        "review_admitted_steps": 55,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": false
      },
      "original_attempts": 1
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "max_steps",
      "controller_wall_seconds": 3353.215531791997,
      "episode_seconds": 539.6389732079988,
      "metering": {
        "maximum_text_input_proxy_tokens": 65214,
        "maximum_reported_text_prompt_tokens": 64198,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0.009892164,
        "native_decision_cost_usd": 0.011262449,
        "actor_text_cost_usd": 2.3082546708,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 2.3195171198,
        "physical_text_responses": 247,
        "reported_completion_tokens": 58515,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 247,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 2.3294092838,
        "physical_responses": 322,
        "application_cache_hits": 274,
        "application_cache_denominator": 596,
        "provider_cache_hit_responses": 242,
        "provider_cache_metadata_responses": 247,
        "provider_cached_input_tokens": 2259968,
        "provider_reported_prompt_tokens": 7453308,
        "native_decision_responses": 75,
        "provider_error_responses": 0,
        "physical_api_seconds": 3098.994288343849,
        "dispatch_wait_seconds": 7.211127414891962,
        "backoff_seconds": 0,
        "observed_callback_seconds": 3123.780666534876
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": null,
          "action_checks": 0,
          "action_matches": 0,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 99,
        "selected_tool_operations": 42,
        "operation_names": {
          "KB_search": 23,
          "get_user_information_by_email": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 2,
          "call_discoverable_agent_tool": 4,
          "grep": 10
        },
        "distinct_exact_tool_operations": 41,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 85,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": true
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "banking_knowledge",
      "task": "task_067",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": 0,
      "score": 0.0,
      "grade_status": "scored",
      "termination": "max_steps",
      "controller_wall_seconds": 345.01158491699607,
      "episode_seconds": 105.26991533397813,
      "metering": {
        "maximum_text_input_proxy_tokens": 65214,
        "maximum_reported_text_prompt_tokens": 64198,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 596,
        "application_cache_denominator": 596,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 3.9105952916725073
      },
      "loop_diagnostics": {
        "original_reward_checks": {
          "db_match": null,
          "action_checks": 0,
          "action_matches": 0,
          "environment_checks": 0,
          "environment_matches": 0
        },
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "historical_policy_materializations": 0,
        "completed_controller_steps": 99,
        "selected_tool_operations": 42,
        "operation_names": {
          "KB_search": 23,
          "get_user_information_by_email": 1,
          "get_current_time": 1,
          "log_verification": 1,
          "unlock_discoverable_agent_tool": 2,
          "call_discoverable_agent_tool": 4,
          "grep": 10
        },
        "distinct_exact_tool_operations": 41,
        "repeated_exact_tool_operations": 1,
        "maximum_no_new_content_streak": 2,
        "last_distinct_source_fragments": 82,
        "review_admitted_steps": 85,
        "format_failure_clarifications": 0,
        "lossless_argument_repairs": 0,
        "premature_original_termination": true
      },
      "controller_recovery": "Only an unscored financial interruption continues after direct human authorization. Same task, seed, models, instructions, original environment and response namespace. Existing unexpired exact responses replay locally; all earlier fees and active attempted wall times remain included. Evaluated rewards are never retried. Approval/coordination hold time is outside active attempted wall.",
      "source_epoch": "unchanged-inference-source-authorized-budget-continuation",
      "original_attempts": 2
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 76.31810279199271,
      "episode_seconds": 76.31810145900818,
      "source_setup_seconds": 13.455493291985476,
      "metering": {
        "maximum_text_input_proxy_tokens": 87260,
        "maximum_reported_text_prompt_tokens": 77562,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0943858152,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0943858152,
        "physical_text_responses": 6,
        "reported_completion_tokens": 1471,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 6,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0943858152,
        "physical_responses": 6,
        "application_cache_hits": 0,
        "application_cache_denominator": 6,
        "provider_cache_hit_responses": 3,
        "provider_cache_metadata_responses": 6,
        "provider_cached_input_tokens": 140032,
        "provider_reported_prompt_tokens": 352910,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 58.57639966611168,
        "dispatch_wait_seconds": 0.18872229193220846,
        "backoff_seconds": 0,
        "observed_callback_seconds": 60.21404096100014
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 1.6443775829975493,
      "episode_seconds": 1.6443759160174523,
      "source_setup_seconds": 1.0420044418424368e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 87260,
        "maximum_reported_text_prompt_tokens": 77562,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 6,
        "application_cache_denominator": 6,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.27622795902425423
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 96.28935033301241,
      "episode_seconds": 84.63658274998306,
      "source_setup_seconds": 9.510408333007945,
      "metering": {
        "maximum_text_input_proxy_tokens": 68681,
        "maximum_reported_text_prompt_tokens": 62729,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.0001621,
        "actor_text_cost_usd": 0.0551725212,
        "other_settled_cost_usd": 6.938893903907228e-18,
        "agent_side_cost_usd": 0.0553346212,
        "physical_text_responses": 6,
        "reported_completion_tokens": 1586,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 6,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0553346212,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 4,
        "provider_cache_metadata_responses": 6,
        "provider_cached_input_tokens": 177792,
        "provider_reported_prompt_tokens": 299982,
        "native_decision_responses": 1,
        "provider_error_responses": 0,
        "physical_api_seconds": 79.51269029104151,
        "dispatch_wait_seconds": 0.1480726249574218,
        "backoff_seconds": 0,
        "observed_callback_seconds": 80.2731910010043
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "frontend"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 5.777615333994618,
      "episode_seconds": 5.771792458021082,
      "source_setup_seconds": 7.080088835209608e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 68681,
        "maximum_reported_text_prompt_tokens": 62729,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.3410000420117285
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "frontend"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 89.11296387499897,
      "episode_seconds": 67.64193320900085,
      "source_setup_seconds": 20.603291500010528,
      "metering": {
        "maximum_text_input_proxy_tokens": 85677,
        "maximum_reported_text_prompt_tokens": 76726,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.075164847,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.075164847,
        "physical_text_responses": 7,
        "reported_completion_tokens": 1195,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 7,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.075164847,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 7,
        "provider_cached_input_tokens": 213120,
        "provider_reported_prompt_tokens": 381759,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 63.32770049897954,
        "dispatch_wait_seconds": 0.03705858302419074,
        "backoff_seconds": 0,
        "observed_callback_seconds": 63.788929624977754
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.5793442919966765,
      "episode_seconds": 1.5725633330002893,
      "source_setup_seconds": 2.9199873097240925e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 85677,
        "maximum_reported_text_prompt_tokens": 76726,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.015854541008593515
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 86.79144991602516,
      "episode_seconds": 74.79434912500437,
      "source_setup_seconds": 11.785345041978871,
      "metering": {
        "maximum_text_input_proxy_tokens": 80448,
        "maximum_reported_text_prompt_tokens": 69771,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0443230272,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0443230272,
        "physical_text_responses": 5,
        "reported_completion_tokens": 944,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0443230272,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 4,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 189952,
        "provider_reported_prompt_tokens": 288384,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 69.90594925099867,
        "dispatch_wait_seconds": 0.02677024999866262,
        "backoff_seconds": 0,
        "observed_callback_seconds": 70.56676441602758
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "product-catalog"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 0.8733393330185208,
      "episode_seconds": 0.8458426669822074,
      "source_setup_seconds": 3.749999450519681e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 80448,
        "maximum_reported_text_prompt_tokens": 69771,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.006698334036627784
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "product-catalog"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 100.62717829100438,
      "episode_seconds": 87.39753370799008,
      "source_setup_seconds": 12.706415832974017,
      "metering": {
        "maximum_text_input_proxy_tokens": 61077,
        "maximum_reported_text_prompt_tokens": 54940,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.0007622600000000001,
        "actor_text_cost_usd": 0.026550913199999998,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0273131732,
        "physical_text_responses": 4,
        "reported_completion_tokens": 1288,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 4,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0273131732,
        "physical_responses": 9,
        "application_cache_hits": 0,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 4,
        "provider_cache_metadata_responses": 4,
        "provider_cached_input_tokens": 122112,
        "provider_reported_prompt_tokens": 179562,
        "native_decision_responses": 5,
        "provider_error_responses": 0,
        "physical_api_seconds": 68.866616541025,
        "dispatch_wait_seconds": 0.022200872946996242,
        "backoff_seconds": 0,
        "observed_callback_seconds": 70.84143220901024
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 2.5864807090256363,
      "episode_seconds": 2.563289915997302,
      "source_setup_seconds": 7.089984137564898e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 61077,
        "maximum_reported_text_prompt_tokens": 54940,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 9,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.01448950098711066
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 64.12211991599179,
      "episode_seconds": 64.12211829100852,
      "source_setup_seconds": 12.143628124991665,
      "metering": {
        "maximum_text_input_proxy_tokens": 63389,
        "maximum_reported_text_prompt_tokens": 58666,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.000621389,
        "actor_text_cost_usd": 0.0176277024,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0182490914,
        "physical_text_responses": 3,
        "reported_completion_tokens": 956,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 3,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0182490914,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 3,
        "provider_cache_metadata_responses": 3,
        "provider_cached_input_tokens": 134784,
        "provider_reported_prompt_tokens": 172280,
        "native_decision_responses": 4,
        "provider_error_responses": 0,
        "physical_api_seconds": 48.35669129199232,
        "dispatch_wait_seconds": 0.013163541007088497,
        "backoff_seconds": 0,
        "observed_callback_seconds": 48.522756248974474
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-4",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 5.412523207982304,
      "episode_seconds": 5.41252225000062,
      "source_setup_seconds": 6.67001586407423e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 63389,
        "maximum_reported_text_prompt_tokens": 58666,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.03198070797952823
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 96.5629947919806,
      "episode_seconds": 76.38576262499555,
      "source_setup_seconds": 19.304749666014686,
      "metering": {
        "maximum_text_input_proxy_tokens": 60570,
        "maximum_reported_text_prompt_tokens": 53832,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.00075258,
        "actor_text_cost_usd": 0.0592970934,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0600496734,
        "physical_text_responses": 4,
        "reported_completion_tokens": 1172,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 4,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0600496734,
        "physical_responses": 9,
        "application_cache_hits": 0,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 1,
        "provider_cache_metadata_responses": 4,
        "provider_cached_input_tokens": 41344,
        "provider_reported_prompt_tokens": 174973,
        "native_decision_responses": 5,
        "provider_error_responses": 0,
        "physical_api_seconds": 63.69560529105365,
        "dispatch_wait_seconds": 0.022639083006652072,
        "backoff_seconds": 0,
        "observed_callback_seconds": 65.14761583300424
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 6.970957833022112,
      "episode_seconds": 6.955406082997797,
      "source_setup_seconds": 5.420006345957518e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 60570,
        "maximum_reported_text_prompt_tokens": 53832,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 9,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.07004116597818211
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 83.36716637498466,
      "episode_seconds": 83.36716420899029,
      "source_setup_seconds": 9.92311362500186,
      "metering": {
        "maximum_text_input_proxy_tokens": 66349,
        "maximum_reported_text_prompt_tokens": 60644,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.000674747,
        "actor_text_cost_usd": 0.0211731036,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0218478506,
        "physical_text_responses": 3,
        "reported_completion_tokens": 1543,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 3,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0218478506,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 3,
        "provider_cache_metadata_responses": 3,
        "provider_cached_input_tokens": 133376,
        "provider_reported_prompt_tokens": 177860,
        "native_decision_responses": 4,
        "provider_error_responses": 0,
        "physical_api_seconds": 69.22629287501331,
        "dispatch_wait_seconds": 0.10239562398055568,
        "backoff_seconds": 0,
        "observed_callback_seconds": 69.71845474894508
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": "execution-error",
      "score": null,
      "grade_status": "execution-error",
      "controller_wall_seconds": 4.502871333010262,
      "episode_seconds": 4.502869500021916,
      "source_setup_seconds": 5.00003807246685e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 66349,
        "maximum_reported_text_prompt_tokens": 60644,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.04173208403517492
      },
      "error_type": "RuntimeError",
      "loop_diagnostics": {
        "scored_root_group_counts": {},
        "execution_failure_reason": "No diagnosis passed the bounded visible JSON and literal-source checks",
        "original_source_queries": null,
        "source_quote_or_output_rejections": null,
        "distinct_valid_hypotheses": null,
        "final_candidates": null,
        "selected_entities_count": null,
        "native_source_controls": null,
        "native_selected_reads": null,
        "denied_repeated_source_queries": null
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 173.22058116699918,
      "episode_seconds": 148.17493716601166,
      "source_setup_seconds": 22.706581915990682,
      "metering": {
        "maximum_text_input_proxy_tokens": 122122,
        "maximum_reported_text_prompt_tokens": 104642,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0500892108,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0500892108,
        "physical_text_responses": 6,
        "reported_completion_tokens": 1055,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 6,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0500892108,
        "physical_responses": 6,
        "application_cache_hits": 0,
        "application_cache_denominator": 6,
        "provider_cache_hit_responses": 6,
        "provider_cache_metadata_responses": 6,
        "provider_cached_input_tokens": 361728,
        "provider_reported_prompt_tokens": 471772,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 142.09727641800418,
        "dispatch_wait_seconds": 0.07377458299743012,
        "backoff_seconds": 0,
        "observed_callback_seconds": 143.5616378740233
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 3.4890152500011027,
      "episode_seconds": 3.4623067080101464,
      "source_setup_seconds": 7.080088835209608e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 122122,
        "maximum_reported_text_prompt_tokens": 104642,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 6,
        "application_cache_denominator": 6,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.0727666660095565
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 280.8584474999807,
      "episode_seconds": 271.4707219169941,
      "source_setup_seconds": 8.63959687500028,
      "metering": {
        "maximum_text_input_proxy_tokens": 130026,
        "maximum_reported_text_prompt_tokens": 108469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.00015512,
        "actor_text_cost_usd": 0.1458915588,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.1460466788,
        "physical_text_responses": 7,
        "reported_completion_tokens": 2241,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 7,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.1460466788,
        "physical_responses": 8,
        "application_cache_hits": 0,
        "application_cache_denominator": 8,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 7,
        "provider_cached_input_tokens": 235008,
        "provider_reported_prompt_tokens": 563964,
        "native_decision_responses": 1,
        "provider_error_responses": 0,
        "physical_api_seconds": 263.31770712096477,
        "dispatch_wait_seconds": 0.028203210065839812,
        "backoff_seconds": 0,
        "observed_callback_seconds": 264.3600591659779
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "product-catalog"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 3.737860458990326,
      "episode_seconds": 3.7338872499822173,
      "source_setup_seconds": 1.0840012691915035e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 130026,
        "maximum_reported_text_prompt_tokens": 108469,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 8,
        "application_cache_denominator": 8,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.02952383307274431
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "product-catalog"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 186.08622883300995,
      "episode_seconds": 175.509865,
      "source_setup_seconds": 8.507271124981344,
      "metering": {
        "maximum_text_input_proxy_tokens": 92135,
        "maximum_reported_text_prompt_tokens": 79117,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0970893594,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0970893594,
        "physical_text_responses": 7,
        "reported_completion_tokens": 1528,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 7,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0970893594,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 7,
        "provider_cached_input_tokens": 213504,
        "provider_reported_prompt_tokens": 431875,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 167.9771147899446,
        "dispatch_wait_seconds": 0.0912221260368824,
        "backoff_seconds": 0,
        "observed_callback_seconds": 169.03674425103236
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.7836413329932839,
      "episode_seconds": 1.7737083330284804,
      "source_setup_seconds": 1.4170072972774506e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 92135,
        "maximum_reported_text_prompt_tokens": 79117,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.019873290992109105
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 82.525155041978,
      "episode_seconds": 69.95513858299819,
      "source_setup_seconds": 11.781968374998542,
      "metering": {
        "maximum_text_input_proxy_tokens": 94706,
        "maximum_reported_text_prompt_tokens": 81293,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0326526312,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0326526312,
        "physical_text_responses": 5,
        "reported_completion_tokens": 727,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0326526312,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 244992,
        "provider_reported_prompt_tokens": 316574,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 58.454059542069444,
        "dispatch_wait_seconds": 0.07830258394824341,
        "backoff_seconds": 0,
        "observed_callback_seconds": 60.104295750003075
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-12",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 1.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.8242214169877116,
      "episode_seconds": 1.7584796250157524,
      "source_setup_seconds": 1.0409858077764511e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 94706,
        "maximum_reported_text_prompt_tokens": 81293,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.06910612402134575
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ConfigMap",
          "name": "flagd-config"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 1,
          "fp": 0,
          "fn": 0
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 88.16246154200053,
      "episode_seconds": 76.05785341598676,
      "source_setup_seconds": 11.621895374992164,
      "metering": {
        "maximum_text_input_proxy_tokens": 99324,
        "maximum_reported_text_prompt_tokens": 80936,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0906773718,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0906773718,
        "physical_text_responses": 5,
        "reported_completion_tokens": 1739,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0906773718,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 2,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 98688,
        "provider_reported_prompt_tokens": 302847,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 73.97219995802152,
        "dispatch_wait_seconds": 0.049504958995385095,
        "backoff_seconds": 0,
        "observed_callback_seconds": 74.79394470900297
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.9959049579920247,
      "episode_seconds": 1.9893227090069558,
      "source_setup_seconds": 8.330098353326321e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 99324,
        "maximum_reported_text_prompt_tokens": 80936,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.06982933598919772
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 55.39396616598242,
      "episode_seconds": 43.566292500006966,
      "source_setup_seconds": 11.07365433301311,
      "metering": {
        "maximum_text_input_proxy_tokens": 116018,
        "maximum_reported_text_prompt_tokens": 89368,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0248719836,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0248719836,
        "physical_text_responses": 4,
        "reported_completion_tokens": 847,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 4,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0248719836,
        "physical_responses": 4,
        "application_cache_hits": 0,
        "application_cache_denominator": 4,
        "provider_cache_hit_responses": 4,
        "provider_cache_metadata_responses": 4,
        "provider_cached_input_tokens": 194176,
        "provider_reported_prompt_tokens": 248052,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 41.836163208965445,
        "dispatch_wait_seconds": 0.01909558300394565,
        "backoff_seconds": 0,
        "observed_callback_seconds": 42.35442999997758
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 5,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.1639972090197261,
      "episode_seconds": 1.142248166986974,
      "source_setup_seconds": 1.4170072972774506e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 116018,
        "maximum_reported_text_prompt_tokens": 89368,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 4,
        "application_cache_denominator": 4,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.021287833020323887
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 5,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 82.27976620802656,
      "episode_seconds": 68.87765033400501,
      "source_setup_seconds": 12.98705933400197,
      "metering": {
        "maximum_text_input_proxy_tokens": 74480,
        "maximum_reported_text_prompt_tokens": 62983,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0547362402,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0547362402,
        "physical_text_responses": 5,
        "reported_completion_tokens": 1963,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0547362402,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 3,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 130432,
        "provider_reported_prompt_tokens": 251257,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 63.53203154198127,
        "dispatch_wait_seconds": 0.025073208002140746,
        "backoff_seconds": 0,
        "observed_callback_seconds": 64.40871008299291
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 4,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 5.637536791997263,
      "episode_seconds": 5.610249541001394,
      "source_setup_seconds": 1.0840012691915035e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 74480,
        "maximum_reported_text_prompt_tokens": 62983,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.08327212595031597
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 4,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 80.90463916701265,
      "episode_seconds": 61.2262173340132,
      "source_setup_seconds": 16.955482207995374,
      "metering": {
        "maximum_text_input_proxy_tokens": 56057,
        "maximum_reported_text_prompt_tokens": 52030,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.000890814,
        "actor_text_cost_usd": 0.0537071196,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.054597933599999995,
        "physical_text_responses": 3,
        "reported_completion_tokens": 1173,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 3,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.054597933599999995,
        "physical_responses": 8,
        "application_cache_hits": 0,
        "application_cache_denominator": 8,
        "provider_cache_hit_responses": 1,
        "provider_cache_metadata_responses": 3,
        "provider_cached_input_tokens": 14336,
        "provider_reported_prompt_tokens": 135336,
        "native_decision_responses": 5,
        "provider_error_responses": 0,
        "physical_api_seconds": 45.388082624966046,
        "dispatch_wait_seconds": 0.07578916600323282,
        "backoff_seconds": 0,
        "observed_callback_seconds": 50.53233837601147
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Node",
          "name": "worker-node-pool"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 5,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 3,
        "denied_repeated_source_queries": 1
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 14.448320416995557,
      "episode_seconds": 14.310942125011934,
      "source_setup_seconds": 1.04197533801198e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 56057,
        "maximum_reported_text_prompt_tokens": 52030,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 8,
        "application_cache_denominator": 8,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.09899158394546248
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Node",
          "name": "worker-node-pool"
        }
      ],
      "initial_distinct_hypotheses": 2,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 5,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 2,
        "final_candidates": 2,
        "selected_entities_count": 1,
        "native_source_controls": 4,
        "native_selected_reads": 3,
        "denied_repeated_source_queries": 1
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 96.23924487500335,
      "episode_seconds": 49.11315370799275,
      "source_setup_seconds": 46.52961158400285,
      "metering": {
        "maximum_text_input_proxy_tokens": 124779,
        "maximum_reported_text_prompt_tokens": 99227,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.056683553399999996,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.056683553399999996,
        "physical_text_responses": 5,
        "reported_completion_tokens": 894,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.056683553399999996,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 252544,
        "provider_reported_prompt_tokens": 378973,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 39.94183954101754,
        "dispatch_wait_seconds": 0.12346412497572601,
        "backoff_seconds": 0,
        "observed_callback_seconds": 41.318165584030794
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.4555317499907687,
      "episode_seconds": 1.4259178749925923,
      "source_setup_seconds": 8.709001122042537e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 124779,
        "maximum_reported_text_prompt_tokens": 99227,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.009663873992394656
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 80.14478270799736,
      "episode_seconds": 46.09138295799494,
      "source_setup_seconds": 33.57486062499811,
      "metering": {
        "maximum_text_input_proxy_tokens": 52724,
        "maximum_reported_text_prompt_tokens": 47656,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.0006181,
        "actor_text_cost_usd": 0.0224675934,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0230856934,
        "physical_text_responses": 5,
        "reported_completion_tokens": 1177,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0230856934,
        "physical_responses": 9,
        "application_cache_hits": 0,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 5,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 137344,
        "provider_reported_prompt_tokens": 185503,
        "native_decision_responses": 4,
        "provider_error_responses": 0,
        "physical_api_seconds": 39.92668291501468,
        "dispatch_wait_seconds": 0.06109058400033973,
        "backoff_seconds": 0,
        "observed_callback_seconds": 41.11940141703235
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-33",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.728331500024069,
      "episode_seconds": 1.7155362920020707,
      "source_setup_seconds": 8.330098353326321e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 52724,
        "maximum_reported_text_prompt_tokens": 47656,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 9,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.017807375959819183
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "cart"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 2,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 127.78799650000292,
      "episode_seconds": 123.0847294999985,
      "source_setup_seconds": 4.277934749989072,
      "metering": {
        "maximum_text_input_proxy_tokens": 77225,
        "maximum_reported_text_prompt_tokens": 70544,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.061593789,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.061593789,
        "physical_text_responses": 5,
        "reported_completion_tokens": 786,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.061593789,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 3,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 163840,
        "provider_reported_prompt_tokens": 302507,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 120.00479550103773,
        "dispatch_wait_seconds": 0.08276062496588565,
        "backoff_seconds": 0,
        "observed_callback_seconds": 121.41445733301225
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "actor-only",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 2.0177765830012504,
      "episode_seconds": 1.982287874998292,
      "source_setup_seconds": 1.1250085663050413e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 77225,
        "maximum_reported_text_prompt_tokens": 70544,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.042473958048503846
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 1,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 1,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 78.66579416600871,
      "episode_seconds": 69.34869474999141,
      "source_setup_seconds": 8.804728249990148,
      "metering": {
        "maximum_text_input_proxy_tokens": 74611,
        "maximum_reported_text_prompt_tokens": 68569,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.0006221,
        "actor_text_cost_usd": 0.0817175586,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0823396586,
        "physical_text_responses": 5,
        "reported_completion_tokens": 1501,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0823396586,
        "physical_responses": 9,
        "application_cache_hits": 0,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 2,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 104576,
        "provider_reported_prompt_tokens": 288565,
        "native_decision_responses": 4,
        "provider_error_responses": 0,
        "physical_api_seconds": 64.21047962200828,
        "dispatch_wait_seconds": 0.031046000018250197,
        "backoff_seconds": 0,
        "observed_callback_seconds": 65.25009779207176
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 3,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 5.483093999995617,
      "episode_seconds": 5.44931437500054,
      "source_setup_seconds": 7.080088835209608e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 74611,
        "maximum_reported_text_prompt_tokens": 68569,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 9,
        "application_cache_denominator": 9,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.06261570903006941
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 3,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 77.27281587501056,
      "episode_seconds": 70.47980500000995,
      "source_setup_seconds": 6.229662625002675,
      "metering": {
        "maximum_text_input_proxy_tokens": 63303,
        "maximum_reported_text_prompt_tokens": 58503,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0.000762724,
        "actor_text_cost_usd": 0.0537258762,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0544886002,
        "physical_text_responses": 3,
        "reported_completion_tokens": 1315,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 3,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0544886002,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 2,
        "provider_cache_metadata_responses": 3,
        "provider_cached_input_tokens": 14592,
        "provider_reported_prompt_tokens": 135349,
        "native_decision_responses": 4,
        "provider_error_responses": 0,
        "physical_api_seconds": 66.22028783295536,
        "dispatch_wait_seconds": 0.06795529101509601,
        "backoff_seconds": 0,
        "observed_callback_seconds": 67.20127137500094
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 4,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "ontology",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 9.889501124998787,
      "episode_seconds": 9.61151770898141,
      "source_setup_seconds": 1.1250085663050413e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 63303,
        "maximum_reported_text_prompt_tokens": 58503,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.05843666600412689
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 6,
        "source_quote_or_output_rejections": 1,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 4,
        "native_selected_reads": 4,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 108.08401125000091,
      "episode_seconds": 92.25268062498071,
      "source_setup_seconds": 15.019673875009175,
      "metering": {
        "maximum_text_input_proxy_tokens": 91844,
        "maximum_reported_text_prompt_tokens": 80380,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.056477646,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.056477646,
        "physical_text_responses": 5,
        "reported_completion_tokens": 906,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 5,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.056477646,
        "physical_responses": 5,
        "application_cache_hits": 0,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 4,
        "provider_cache_metadata_responses": 5,
        "provider_cached_input_tokens": 191360,
        "provider_reported_prompt_tokens": 317798,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 89.60313333396334,
        "dispatch_wait_seconds": 0.04411000001709908,
        "backoff_seconds": 0,
        "observed_callback_seconds": 90.63359316700371
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "actor-only",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 3.1276667089841794,
      "episode_seconds": 3.0985466669953894,
      "source_setup_seconds": 1.4170072972774506e-06,
      "metering": {
        "maximum_text_input_proxy_tokens": 91844,
        "maximum_reported_text_prompt_tokens": 80380,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 5,
        "application_cache_denominator": 5,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.08151558498502709
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 8,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 204.2397644170269,
      "episode_seconds": 192.89696020900737,
      "source_setup_seconds": 10.01113087500562,
      "metering": {
        "maximum_text_input_proxy_tokens": 106890,
        "maximum_reported_text_prompt_tokens": 91741,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.1166737704,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.1166737704,
        "physical_text_responses": 7,
        "reported_completion_tokens": 2319,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 7,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.1166737704,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 6,
        "provider_cache_metadata_responses": 7,
        "provider_cached_input_tokens": 225664,
        "provider_reported_prompt_tokens": 487374,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 185.91407875099685,
        "dispatch_wait_seconds": 0.14460425099241547,
        "backoff_seconds": 0,
        "observed_callback_seconds": 188.47664083301788
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "perplexity/pplx-decider-v1.1-27b",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 1.078950374998385,
      "episode_seconds": 1.0581532920186874,
      "source_setup_seconds": 7.089984137564898e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 106890,
        "maximum_reported_text_prompt_tokens": 91741,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.07184858398977667
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "primary",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 89.79505529100425,
      "episode_seconds": 83.93902691701078,
      "source_setup_seconds": 4.908786291984143,
      "metering": {
        "maximum_text_input_proxy_tokens": 93219,
        "maximum_reported_text_prompt_tokens": 82536,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0.0832529028,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0.0832529028,
        "physical_text_responses": 7,
        "reported_completion_tokens": 1547,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 7,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0.0832529028,
        "physical_responses": 7,
        "application_cache_hits": 0,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 7,
        "provider_cache_metadata_responses": 7,
        "provider_cached_input_tokens": 274048,
        "provider_reported_prompt_tokens": 460072,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 80.20062499694177,
        "dispatch_wait_seconds": 0.08521570797893219,
        "backoff_seconds": 0,
        "observed_callback_seconds": 81.68988974802778
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    },
    {
      "domain": "ITBench",
      "task": "Scenario-102",
      "seed": 613,
      "arm": "cloudflare/clef-flash",
      "profile": "reference",
      "phase": "warm",
      "status": "prediction-complete",
      "score": 0.0,
      "grade_status": "scored",
      "controller_wall_seconds": 3.654179750010371,
      "episode_seconds": 3.631976292002946,
      "source_setup_seconds": 7.920025382190943e-07,
      "metering": {
        "maximum_text_input_proxy_tokens": 93219,
        "maximum_reported_text_prompt_tokens": 82536,
        "text_input_proxy_limit": 262144,
        "simulator_cost_usd": 0,
        "native_decision_cost_usd": 0,
        "actor_text_cost_usd": 0,
        "other_settled_cost_usd": 0,
        "agent_side_cost_usd": 0,
        "physical_text_responses": 0,
        "reported_completion_tokens": 0,
        "reported_reasoning_tokens": 0,
        "reasoning_metadata_responses": 0,
        "reasoning_metadata_exceeds_completion_responses": 0,
        "actual_cost_usd": 0,
        "physical_responses": 0,
        "application_cache_hits": 7,
        "application_cache_denominator": 7,
        "provider_cache_hit_responses": 0,
        "provider_cache_metadata_responses": 0,
        "provider_cached_input_tokens": 0,
        "provider_reported_prompt_tokens": 0,
        "native_decision_responses": 0,
        "provider_error_responses": 0,
        "physical_api_seconds": 0,
        "dispatch_wait_seconds": 0,
        "backoff_seconds": 0,
        "observed_callback_seconds": 0.08271537508699112
      },
      "submitted_entities": [
        {
          "namespace": "otel-demo",
          "kind": "Deployment",
          "name": "ad"
        },
        {
          "namespace": "otel-demo",
          "kind": "ResourceQuota",
          "name": "otel-demo-memory"
        }
      ],
      "initial_distinct_hypotheses": 1,
      "loop_diagnostics": {
        "scored_root_group_counts": {
          "tp": 0,
          "fp": 2,
          "fn": 1
        },
        "execution_failure_reason": null,
        "original_source_queries": 10,
        "source_quote_or_output_rejections": 0,
        "distinct_valid_hypotheses": 1,
        "final_candidates": 1,
        "selected_entities_count": 2,
        "native_source_controls": 0,
        "native_selected_reads": 0,
        "denied_repeated_source_queries": 0
      },
      "original_attempts": 1
    }
  ],
  "execution_schedule": {
    "maximum_parallel_conditions": 3,
    "description": "Original counterbalanced eight-task cohort retained. Three valid reference primary/warm pairs are reused unchanged from the earlier source epoch, including evaluated zeros. The three unscored ontology source-index failures continue in fresh namespaces with all original attempts, fees and wall times retained. Only implicit diagnostic source catalogs are batched; actual action bounds and model instructions are unchanged. Remaining case/profile blocks admit all three arms concurrently. Immediate warm attempts issue no physical inference. Times include actual provider queueing and environment initialization.",
    "initial_cohort_fixed_before_primary": true,
    "source_repair_amendment": {
      "at_utc": "2026-10-16T04:07:17.149999+00:00",
      "kind": "bounded-diagnostic-source-catalog",
      "observed_failure": "Original bank source indexing mistakenly submitted more than32diagnostic source-read entries as a single action catalog. All3ontology arms failed without an official reward.",
      "change": "Batch only implicit diagnostic source catalogs in groups of32. Keep the real action-catalog32limit, source facts/revision, candidate rules, text/native settings and all model instructions unchanged.",
      "existing_results": "Retain the three valid reference primary/warm results exactly. Reference code never invokes the changed ontology bridge. Never retry their evaluated zeros.",
      "continuation": "Fresh cold namespaces for the3unscored ontology conditions only, with all original failed primary/warm attempts, fees and attempted wall time retained in attempt_history. No cached model response imported from the failed namespace. Continue the same original preselected8task cohort; no outcome selection or new labels.",
      "qualification": "Original six known-case ontology controls (Actor, Decider, Clef in both domains), plus offline source-count tests at32,33,256 and unchanged real-action32guard. No win/score threshold."
    },
    "financial_continuation": {
      "authorization": {
        "schema": "mithril.mimo-budget-amendment/v1",
        "prior_total_cap_usd": 60,
        "authorized_total_cap_usd": 65,
        "authorized_at_utc": "2026-10-16T05:54:54.985088+00:00",
        "human_instruction_sha256": "aedd392fca448f997fda7fbcc55b8fc5968920186576fe67f841400202276518",
        "source": "Direct human authorization; increase used only as required to complete the existing comparison.",
        "same_inference_source": true,
        "same_preselected_matrix": true,
        "same_models_and_settings": true,
        "prior_invoices_and_uncertain_reserves_preserved": true,
        "prior_confirmed_usd": 58.81791458520016,
        "prior_uncertain_reserve_usd": 0.9812594200000045,
        "boundary": "Only previously unscored financial interruptions and unattempted conditions continue. Evaluated rewards, including zeros, are preserved. No automatic increase above USD65."
      },
      "original_prediction_manifest_sha256": "0c9f79c03e8e2f27c89b8596c06128d1a99ef920cab9cafe8e35a697028ed238",
      "same_inference_source": true,
      "same_original_matrix": true,
      "same_maximum_parallel_conditions": 3,
      "cache_policy": "Continue only unscored financial interruptions in the same original response namespace. Unexpired exact responses replay locally; tools execute fresh in the original environment. All original fees and active attempted walls are retained. Approval/coordination holds are excluded from active-attempt timing. Valid rewards, including zeros, are never retried."
    },
    "prospective_serial_IT_admission": {
      "schema": "mithril.mimo-admission-amendment/v1",
      "recorded_at_utc": "2026-10-16T07:29:47.369637+00:00",
      "recorded_before_selected_IT_inference": true,
      "original_maximum_parallel_conditions": 3,
      "remaining_maximum_parallel_conditions": 1,
      "authorized_total_cap_usd": 70,
      "human_budget_amendment_sha256": "fe8a73d669065ac97b08b8dc25e5c6d8be7d1060e8e955e3749a4ae29e4a2df7",
      "reason": "Reduce simultaneous unsettled reservations under the unchanged human USD70 ceiling. All selected IT arms receive the same one-condition admission.",
      "same_original_matrix_and_arm_order": true,
      "same_inference_source": true,
      "same_models_inputs_seeds_tools_and_limits": true,
      "worker_lifecycle": "Fresh original per-condition Python worker for every primary; its immediate no-inference warm attempt shares that worker and source reader, as in the initial original matrix.",
      "recovery_order": "Within an already attempted bank block, financial interruptions continue before transient payload failures to preserve the large exact-response prefix. Unattempted IT order remains the original rotating order.",
      "financial_recovery": "Only explicit unscored financial interruptions or one proven provider 429/5xx payload transport error may continue in the same response namespace. Every prior fee and active-attempt wall is retained; valid rewards, including zeros, are never retried.",
      "timing_boundary": "Earlier bank blocks admitted up to three conditions. Remaining IT conditions admit one. These are observed end-to-end times under disclosed admission, not isolated model-speed or cross-domain estimates. No in-flight measurement was paused."
    }
  }
}