{
  "not_run": [
    {
      "needs": "a live local or frontier endpoint",
      "standing_result": "retired on 2026-07-26. The arms were not independent: the treatment's first attempt is the same call as the baseline's only attempt, so the treatment cannot score lower and the difference is not a comparison. The quantity measured is verified pass@k. The retired table read verified inference 9/10 against single-shot 8/10, difference +0.100 with 95% CI [-0.236, +0.420], an interval that includes zero, and no capability uplift is claimed.",
      "suite": "m7 capability arms",
      "where": "handoff/site-designer/BENCHMARKS.md, docs/claims/2026-07-13-uplift"
    },
    {
      "needs": "a provider list and an oracle",
      "standing_result": null,
      "suite": "uplift_bench paired arms",
      "where": "harness/uplift_bench.py, POST /api/uplift"
    },
    {
      "needs": "endpoints and a private task set the operator supplies",
      "standing_result": null,
      "suite": "verified_bench private task set",
      "where": "harness/verified_bench.py, POST /api/bench/run"
    },
    {
      "needs": "a chat backend per mode",
      "standing_result": null,
      "suite": "classifier friction backend modes",
      "where": "harness/classifier_friction_bench.py"
    },
    {
      "needs": "a chat backend; the deterministic variants below ran instead",
      "standing_result": null,
      "suite": "backend variants of the governed, recovery, stateful and source-mined suites",
      "where": "run_backend_* in the same modules"
    }
  ],
  "parity": {
    "absent": 0,
    "declared_on": "2026-09-03",
    "gaps": [],
    "rows": 33,
    "uniquely_witnessed": 24,
    "witnessed": 33
  },
  "python": "3.12.10",
  "result_sha256": "107a33f920629afb53ddaf0d960ea54f3bed96d6c21c9ed99a0acecd2ac1579b",
  "schema": "flywheel.offline-benchmarks/v1",
  "suites": [
    {
      "credible": true,
      "detail": [
        {
          "name": "re_checkability",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "externalization",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "adversarial_soundness",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "no_regression",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "invariant_fidelity",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "null_space_honesty",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "provenance",
          "score": 1.0,
          "strawman": 0.0
        },
        {
          "name": "buildc_receipt_bridge",
          "score": 1.0,
          "strawman": null
        }
      ],
      "headline": {
        "dimensions": 8,
        "harness_overall": 1.0,
        "separation": 1.0,
        "strawman_overall": 0.0
      },
      "name": "accountability",
      "non_goal": "measures accountability, NOT capability. Pair it with a capability bench",
      "question": "does an unaccountable system score badly here",
      "schema": "accountability/v1",
      "seconds": 8.008
    },
    {
      "detail": [
        {
          "name": "failed_cases",
          "score": 0
        },
        {
          "name": "mean_autonomy_promotion_evidence_score",
          "score": 0.5
        },
        {
          "name": "mean_docs_schematic_drift_score",
          "score": 1.0
        },
        {
          "name": "mean_execution_graph_coverage",
          "score": 1.0
        },
        {
          "name": "mean_fitness_metric_coverage",
          "score": 0.167
        },
        {
          "name": "mean_hitl_gate_score",
          "score": 0.667
        },
        {
          "name": "mean_maturity_gate_score",
          "score": 0.833
        },
        {
          "name": "mean_organic_doc_update_score",
          "score": 1.0
        },
        {
          "name": "mean_quality_score",
          "score": 0.542
        },
        {
          "name": "mean_receipt_auditability_score",
          "score": 1.0
        },
        {
          "name": "mean_sandboxed_skill_mutation_score",
          "score": 0.167
        },
        {
          "name": "mean_source_of_truth_score",
          "score": 1.0
        },
        {
          "name": "mean_ui_state_grounding_score",
          "score": 0.167
        },
        {
          "name": "mean_vector_acceleration_boundary",
          "score": 0.167
        },
        {
          "name": "pass_rate",
          "score": 1.0
        },
        {
          "name": "passed_cases",
          "score": 6
        },
        {
          "name": "unauthorized_write_count",
          "score": 0
        },
        {
          "name": "unsafe_mutation_count",
          "score": 0
        }
      ],
      "headline": {
        "failed": 0,
        "mean_quality_score": 0.542,
        "pass_rate": 1.0,
        "passed": 6,
        "scenarios": 6
      },
      "name": "governed-agent",
      "question": "does a workflow refuse an action above its tier",
      "schema": "governed-agent-workflow-benchmark/v1",
      "seconds": 0.011
    },
    {
      "detail": [
        {
          "name": "avg_attempts",
          "score": 1.5
        },
        {
          "name": "avg_retries",
          "score": 0.5
        },
        {
          "name": "fallback_quality",
          "score": 0.958
        },
        {
          "name": "fallback_use_rate",
          "score": 0.167
        },
        {
          "name": "max_recovery_latency",
          "score": 1745
        },
        {
          "name": "mean_recovery_latency",
          "score": 756.667
        },
        {
          "name": "p95_recovery_latency",
          "score": 1745
        },
        {
          "name": "receipt_completeness",
          "score": 1.0
        },
        {
          "name": "recovery_success_rate",
          "score": 1.0
        },
        {
          "name": "retry_budget_compliance",
          "score": 1.0
        },
        {
          "name": "retry_use_rate",
          "score": 0.5
        },
        {
          "name": "scenario_fail_count",
          "score": 0
        },
        {
          "name": "scenario_pass_count",
          "score": 6
        },
        {
          "name": "silent_failure_rate",
          "score": 0.0
        },
        {
          "name": "task_correct_rate",
          "score": 0.833
        },
        {
          "name": "typed_escalation_rate",
          "score": 0.167
        }
      ],
      "headline": {
        "receipt_completeness": 1.0,
        "recovery_success_rate": 1.0,
        "scenarios": 6,
        "silent_failure_rate": 0.0
      },
      "name": "agent-recovery",
      "question": "does an injected fault recover without failing quietly",
      "schema": "agent.recovery-benchmark/v1",
      "seconds": 0.042
    },
    {
      "detail": [
        {
          "name": "correction_permanence_score",
          "score": 1.0
        },
        {
          "name": "negative_ledger_enforcement",
          "score": 1.0
        },
        {
          "name": "persistent_memory_replay_score",
          "score": 1.0
        },
        {
          "name": "teacher_exit_evidence_score",
          "score": 1.0
        },
        {
          "name": "selfplay_nonreinforcement_score",
          "score": 1.0
        },
        {
          "name": "fixed_item_scorecard_reproducibility",
          "score": 1.0
        },
        {
          "name": "negative_result_preservation",
          "score": 1.0
        },
        {
          "name": "discord_scope_lock_score",
          "score": 1.0
        },
        {
          "name": "tool_trace_receipt_completeness",
          "score": 1.0
        },
        {
          "name": "secret_boundary_score",
          "score": 1.0
        }
      ],
      "headline": {
        "checks": 10,
        "pass_rate": 1.0,
        "passed": true
      },
      "name": "stateful-provider-swap",
      "question": "does state survive a provider swap",
      "schema": "unisonai.stateful-benchmark/v1",
      "seconds": 0.006
    },
    {
      "detail": [],
      "headline": {
        "cases": 26,
        "failed": 0,
        "metrics_asserted": 170,
        "pass_rate": 1.0,
        "passed": 26
      },
      "name": "source-mined",
      "question": "do the mined checks still hold against their datasets",
      "schema": "source-mined.executable-benchmark/v1",
      "seconds": 0.014
    },
    {
      "caveats": [
        "The model references in these artifacts no longer resolve to anything on a live roster, so the exact weights behind each arm cannot be fetched again from the reference alone. The per-task outcomes are committed and the arithmetic here is reproducible; the generation is not.",
        "One deterministic sample per task per arm. A single sample measures the greedy decode, not the model.",
        "This compares base weights against continued-pretrained weights on general code completion. It is not a measurement of the verification harness, which is the thing this repository is."
      ],
      "detail": [],
      "headline": {
        "delta_points": -0.0305,
        "gains": 9,
        "p_exact": 0.4049,
        "regressions": 14,
        "tasks": 164
      },
      "name": "paired-replication",
      "non_goal": "continued pretraining on the workspace corpus did not improve general code completion and did not significantly harm it; the point estimate is a regression inside the noise",
      "question": "did continued pretraining change general code completion",
      "schema": "flywheel.paired-replication/v1",
      "seconds": 0.001
    }
  ]
}