{
  "schema_version": "bash-sft-02-fair-eval-comparison-v1",
  "updated_date": "2026-08-21",
  "components": 98,
  "bootstrap": {
    "paired_by_component": true,
    "resamples": 10000,
    "seed": 42,
    "interval": 0.95
  },
  "comparisons": {
    "original_strict_reference": {
      "stock": {
        "passes": 0,
        "n": 98,
        "pass_rate": 0.0,
        "raw_command_contract": 0
      },
      "arm2": {
        "passes": 88,
        "n": 98,
        "pass_rate": 0.897959,
        "raw_command_contract": 98
      },
      "interpretation": "response-contract-confounded; historical reference only"
    },
    "symmetric_whole_response_markdown_adapter": {
      "stock": {
        "passes": 9,
        "n": 98,
        "pass_rate": 0.091837,
        "raw_command_contract": 0,
        "adapter_accepted": 98,
        "policy_accepted": 69
      },
      "prior_session_2": {
        "passes": 10,
        "n": 98,
        "pass_rate": 0.102041,
        "raw_command_contract": 46,
        "adapter_accepted": 97,
        "policy_accepted": 84
      },
      "arm2": {
        "passes": 88,
        "n": 98,
        "pass_rate": 0.897959,
        "raw_command_contract": 98,
        "adapter_accepted": 98,
        "policy_accepted": 98
      },
      "paired_arm2_minus_stock": {
        "pass_count_delta": 79,
        "pass_rate_delta": 0.806122,
        "ci95_low": 0.714286,
        "ci95_high": 0.887755
      }
    },
    "identical_contract_calibrated_generation": {
      "stock": {
        "passes": 8,
        "n": 98,
        "pass_rate": 0.081633,
        "raw_command_contract": 98,
        "policy_accepted": 86,
        "syntax_valid": 85,
        "executed": 85
      },
      "arm2": {
        "passes": 89,
        "n": 98,
        "pass_rate": 0.908163,
        "raw_command_contract": 98,
        "policy_accepted": 98,
        "syntax_valid": 98,
        "executed": 98
      },
      "paired_arm2_minus_stock": {
        "pass_count_delta": 81,
        "pass_rate_delta": 0.826531,
        "ci95_low": 0.744898,
        "ci95_high": 0.897959
      }
    }
  },
  "conclusion": "both fair conditions show a low non-zero stock baseline and an approximately 81 to 83 percentage-point Arm 2 advantage on the within-generator set"
}
