{
  "artifact": "s08_parity_bar",
  "sprint": "S08-O4",
  "prereg": "results/s08_parity_bar.prereg.json",
  "claim": "This lane's stranger verifier is graded against its own declared rule set on five clauses, with both degenerate strategies failing the bar first.",
  "method": "tools/parity_bar.py --manifest configs/parity_bar_vllm.json; mutants are edits to committed certificates, listed with before/after in C4.",
  "manifest_used": {
    "lane": "vllm",
    "repo": "..",
    "verifier": {
      "path": "verifier/verify_core.py",
      "verify": "verify",
      "verdict": "verdict_of"
    },
    "bundle_pairs": [
      [
        "verifier/bundle/stats.py",
        "bench/stats.py"
      ],
      [
        "verifier/bundle/seal.py",
        "vendor/seal.py"
      ],
      [
        "verifier/bundle/economics.py",
        "bench/economics.py"
      ],
      [
        "verifier/bundle/vendor/eprocess.py",
        "vendor/eprocess.py"
      ],
      [
        "verifier/bundle/vendor/__init__.py",
        "vendor/__init__.py"
      ]
    ],
    "entry_point": {
      "file": "verifier/index.html",
      "regex": "const FILES\\s*=\\s*\\[(.*?)\\]",
      "root": "verifier",
      "import": "verify_core",
      "extra_sys_path": [
        "bundle"
      ]
    },
    "certificates": {
      "glob": "results/*.json",
      "exclude_suffixes": [
        ".raw.json"
      ]
    },
    "rules": [
      {
        "id": "seal_recomputes",
        "check_prefix": "seal",
        "mutant": {
          "certificate": "results/s46c_apc_fixed_bridge_n8.prereg.json",
          "path": [
            "threshold",
            "paired_cv_pct_max"
          ],
          "op": "perturb_number"
        },
        "forgery": "a threshold moved after it was sealed"
      },
      {
        "id": "mde_recomputes",
        "check_prefix": "MDE",
        "mutant": {
          "certificate": "results/s01_aa_null_sharded.json",
          "path": [
            "analysis",
            "gpu_seconds",
            "mde_table",
            0,
            "mde_pct"
          ],
          "op": "perturb_number"
        },
        "forgery": "a detectable-effect figure that does not come out of the stated method"
      },
      {
        "id": "replicate_floor",
        "check_prefix": "no band below",
        "mutant": {
          "certificate": "results/s02_paired_knee.json",
          "path": [
            "paired_knee_analysis",
            "n_replicates"
          ],
          "op": "set_number:3"
        },
        "forgery": "a confidence band quoted below the minimum replicates"
      },
      {
        "id": "economics_identity",
        "check_prefix": "$/1M",
        "mutant": {
          "certificate": "results/sprint44_knee_claim_path.json",
          "path": [
            "compare",
            "equal_gpu_hours_within_tolerance"
          ],
          "op": "set_true"
        },
        "forgery": "a dollar figure presented as evidence at equal GPU-hours, where it is an identity"
      }
    ],
    "no_op_mutant": {
      "certificate": "results/s06_c1_fixed_bridge_n8.json",
      "path": [
        "config",
        "reps"
      ],
      "op": "set_number:999",
      "why": "no rule reads config.reps; the declared scope says so and this control keeps it honest"
    }
  },
  "lane": "vllm",
  "manifest": "configs/parity_bar_vllm.json",
  "C0_degenerate_controls": {
    "accept_all": {
      "bar_passes": false,
      "C3_disagreements": 0,
      "C4_rules_tripped_by_their_mutant": 0,
      "C5_silent": true,
      "must_fail_the_bar": true,
      "fails_the_bar_as_required": true
    },
    "reject_all": {
      "bar_passes": false,
      "C3_disagreements": 142,
      "C4_rules_tripped_by_their_mutant": 0,
      "C5_silent": true,
      "must_fail_the_bar": true,
      "fails_the_bar_as_required": true
    }
  },
  "C0_pass": true,
  "C1_bundle_identical": {
    "pairs": [
      {
        "bundled": "verifier/bundle/stats.py",
        "source": "bench/stats.py",
        "bundled_sha256": "0a7489b1f4690642da74c86b78a704eaf4af0f94c15d0d8cfb2a7e12348604b4",
        "source_sha256": "0a7489b1f4690642da74c86b78a704eaf4af0f94c15d0d8cfb2a7e12348604b4",
        "identical": true
      },
      {
        "bundled": "verifier/bundle/seal.py",
        "source": "vendor/seal.py",
        "bundled_sha256": "7ceb89234f1929c4a86fadd7f595b9cca1c5129539f91a96e78140a3bc2b2077",
        "source_sha256": "7ceb89234f1929c4a86fadd7f595b9cca1c5129539f91a96e78140a3bc2b2077",
        "identical": true
      },
      {
        "bundled": "verifier/bundle/economics.py",
        "source": "bench/economics.py",
        "bundled_sha256": "0e24521370a2a7b64102bc9f54fe93a82368884b16d73f47c5d63bdbd9576d62",
        "source_sha256": "0e24521370a2a7b64102bc9f54fe93a82368884b16d73f47c5d63bdbd9576d62",
        "identical": true
      },
      {
        "bundled": "verifier/bundle/vendor/eprocess.py",
        "source": "vendor/eprocess.py",
        "bundled_sha256": "12727ba38794000a13d135cdb8d6dbdafa2674381ef34a8a26334d7a7849c381",
        "source_sha256": "12727ba38794000a13d135cdb8d6dbdafa2674381ef34a8a26334d7a7849c381",
        "identical": true
      },
      {
        "bundled": "verifier/bundle/vendor/__init__.py",
        "source": "vendor/__init__.py",
        "bundled_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
        "source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
        "identical": true
      }
    ],
    "drifted": [],
    "pass": true
  },
  "C2_bundle_self_contained": {
    "entry_point": "verifier/index.html",
    "files_it_fetches": [
      "bundle/stats.py",
      "bundle/seal.py",
      "bundle/economics.py",
      "bundle/vendor/__init__.py",
      "bundle/vendor/eprocess.py",
      "verify_core.py"
    ],
    "files_missing_from_repo": [],
    "imports_alone": true,
    "error": "",
    "pass": true
  },
  "C3_recompute": {
    "certificates_graded": 27,
    "checks": 115,
    "disagreements": 28,
    "disagreeing": [
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (gpu_seconds, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (gpu_seconds, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=10, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (output_tok_s, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_aa_null_tiny.json",
        "rule": "MDE recomputes (output_tok_s, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (gpu_seconds, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (gpu_seconds, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=10, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=10, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control.json",
        "rule": "MDE recomputes (output_tok_s, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (gpu_seconds, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (gpu_seconds, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (ttft_p95_ms, n=10, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (tpot_p95_ms, n=10, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (output_tok_s, n=3, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      },
      {
        "certificate": "results/s01_positive_control_eager.json",
        "rule": "MDE recomputes (output_tok_s, n=5, exact noncentral-t)",
        "verdict": "FAIL: the MDE does not recompute"
      }
    ],
    "n_disagreeing_listed": 28
  },
  "C4_rule_coverage": {
    "rules": [
      {
        "rule_id": "seal_recomputes",
        "forgery": "a threshold moved after it was sealed",
        "committed_certificates_exercising_it": 13,
        "committed_certificates_where_it_FAILS": 0,
        "corpus_exercises_the_refusing_direction": false,
        "mutant": {
          "certificate": "results/s46c_apc_fixed_bridge_n8.prereg.json",
          "path": [
            "threshold",
            "paired_cv_pct_max"
          ],
          "op": "perturb_number",
          "before": 8.72,
          "after": 18.44
        },
        "mutant_trips_the_rule": true
      },
      {
        "rule_id": "mde_recomputes",
        "forgery": "a detectable-effect figure that does not come out of the stated method",
        "committed_certificates_exercising_it": 8,
        "committed_certificates_where_it_FAILS": 3,
        "corpus_exercises_the_refusing_direction": true,
        "mutant": {
          "certificate": "results/s01_aa_null_sharded.json",
          "path": [
            "analysis",
            "gpu_seconds",
            "mde_table",
            0,
            "mde_pct"
          ],
          "op": "perturb_number",
          "before": 0.029,
          "after": 1.058
        },
        "mutant_trips_the_rule": true
      },
      {
        "rule_id": "replicate_floor",
        "forgery": "a confidence band quoted below the minimum replicates",
        "committed_certificates_exercising_it": 5,
        "committed_certificates_where_it_FAILS": 0,
        "corpus_exercises_the_refusing_direction": false,
        "mutant": {
          "certificate": "results/s02_paired_knee.json",
          "path": [
            "paired_knee_analysis",
            "n_replicates"
          ],
          "op": "set_number:3",
          "before": 5,
          "after": 3
        },
        "mutant_trips_the_rule": true
      },
      {
        "rule_id": "economics_identity",
        "forgery": "a dollar figure presented as evidence at equal GPU-hours, where it is an identity",
        "committed_certificates_exercising_it": 1,
        "committed_certificates_where_it_FAILS": 0,
        "corpus_exercises_the_refusing_direction": false,
        "mutant": {
          "certificate": "results/sprint44_knee_claim_path.json",
          "path": [
            "compare",
            "equal_gpu_hours_within_tolerance"
          ],
          "op": "set_true",
          "before": false,
          "after": true
        },
        "mutant_trips_the_rule": true
      }
    ],
    "pass": true,
    "rules_no_committed_certificate_exercises": [],
    "rules_never_refusing_in_the_corpus": [
      "seal_recomputes",
      "replicate_floor",
      "economics_identity"
    ]
  },
  "C5_no_op_mutant_silent": {
    "certificate": "results/s06_c1_fixed_bridge_n8.json",
    "path": [
      "config",
      "reps"
    ],
    "op": "set_number:999",
    "before": null,
    "after": 999,
    "why": "no rule reads config.reps; the declared scope says so and this control keeps it honest",
    "failures_before": 0,
    "failures_after": 0,
    "silent": true,
    "pass": true
  },
  "bar_passes": true,
  "bar_met": true,
  "verdict": "BAR MET for lane vllm: controls both fail as required; bundle 5/5 identical and imports alone; 4/4 rules tripped by their mutant; 0 rule(s) no committed certificate exercises (none); 3 rule(s) the corpus never exercises in the refusing direction (seal_recomputes, replicate_floor, economics_identity); C3 reported: 28 disagreement(s) over 115 checks on 27 certificates.",
  "wall_s": 12.7,
  "cost_usd": 0.0,
  "does_not_prove": [
    "That the rules are the right rules: the bar grades a verifier against the rule set it declares, so a lane declaring one weak rule and mutating it passes C4 with one weak rule.",
    "That a browser runs it: C2 imports the entry point's file list under CPython, which is strictly more generous than pyodide.",
    "Anything about foreign certificate formats: that is the cross-verification matrix, run and closed by the orchestrator on 2026-09-04.",
    "That C3's disagreements are the verifier's fault: they are read from committed certificates, some written before the exact-MDE fix."
  ],
  "generated_utc": "2026-09-04T19:58:42.548855+00:00",
  "git_rev": "fd54c78",
  "git_dirty": true,
  "run_id": "27a8e3b6e2b941b6",
  "host_kind": "Darwin-arm64",
  "provenance_schema": "gpu-cert-provenance/1"
}