{
  "id": "sigma-corpus-audit",
  "item": "SN-1 — name and freeze the σ corpus behind the credibility claim, and fix its unit of replication",
  "measured": "2026-08-17T01:54:54.772Z",
  "recomputeWith": "node scripts/pmu/sigma-corpus-audit.mjs",
  "repo": "thetadrivencoach",
  "headCommit": "08a7b072c848995d3a7c062d195b37c6445bf893",
  "contract": {
    "rule": "every published σ headline READS FROM this file — corpus id, n, the quantile it quotes, and the digest under which it was frozen. A σ headline that names no corpus id is not publishable.",
    "recomputability": "run the recomputeWith command on the same source bytes and the digests below reproduce; a changed digest means the corpus moved, and the headline is stale until re-frozen.",
    "threeCorporaAreNotOneCorpus": "commit-page σ, lens σ and drift σ answer different questions with different nulls. No headline may pool them, and a headline that does not say which one it used is unresolvable."
  },
  "corpora": {
    "commitPages": {
      "id": "commit-page-sigma",
      "sourceGlob": "public/commit/<sha>/index.html",
      "sigmaField": "the rendered \"σ (placement measurement)\" row — renderAttestation({ sigma })",
      "producedBy": "scripts/pmu/auditor-attestation.mjs (renderAttestation, the σ row) ← commit-triptych meta.matchSigma — the two walks' cloud overlap scored against impostors",
      "filterPredicate": "pages matching /<b>NUM</b> <span>shape-match &sigma;</span>/ ; pages carrying only a 0..1 \"shape-match fit\" (a delegation-QC confidence, a DIFFERENT measure) or the \"no walk (unplaced / churn)\" placeholder are EXCLUDED and counted",
      "unitOfReplication": {
        "clusterKeyAvailable": false,
        "reason": "the commit page records no reef/lens/corpus sha, so a cluster key cannot be derived from this surface alone",
        "remedy": "join by commit sha to data/pmu/measure-history*.ndjson (done below in the joined block) — only the joined subset carries a defensible cluster key",
        "n_raw": 3554,
        "n_distinct": 3554,
        "claimFence": "a rate over these readings is POOLED across commits that share most of their reality corpus, the reef and the snippet library — it may be reported as a pooled rate over commit pages, and must not be reported as a rate over independent observations"
      },
      "present": true,
      "n_candidates": 3598,
      "admitted": 3554,
      "excluded": {
        "missingIndexHtml": 1,
        "fitButNoSigma": 38,
        "noWalkPlaceholder": 0,
        "unrecognised": 5,
        "unrecognisedSample": [
          "3dc3d72a5",
          "6a7c55dc5",
          "75356bd38",
          "b582eb3c4",
          "c9a7fab3e"
        ]
      },
      "distribution": {
        "n": 3554,
        "min": -5.43,
        "max": 55.37,
        "mean": 1.9336,
        "quantiles": {
          "p05": -2.51,
          "p25": -0.9675,
          "median": 0,
          "p75": 0.9675,
          "p95": 20.0375
        },
        "aboveZero": 1188,
        "aboveZeroPct": 33.43,
        "positivesOnly": {
          "n": 1188,
          "median": 2.575,
          "max": 55.37
        }
      },
      "digests": {
        "readingsSha": "ab93162d57069295d3e004d67d045db1ffb69ff54aa844b3f55cfc7666bbd364",
        "bytesSha": "802e76994727e0d8decebd6b9f067f62d98bc41117ef5aae66b77d8206b4eca3"
      }
    },
    "lensReceipts": {
      "id": "lens-receipt-sigma",
      "sourceGlob": ".thetacog/lens-receipts/*.json",
      "sigmaField": "sigma",
      "producedBy": "scripts/pmu/prompt-lens.mjs → result.boundary.sigma — INTENT-PLACEMENT from the lens walk. prompt-lens.mjs:819 states in the source that this is \"NOT the commit's coverage σ\".",
      "filterPredicate": "every *.json whose top-level `sigma` parses as a finite number",
      "notSameAs": "commit-page-sigma — a different question (where did this prompt place?) against a different null. Pooling the two is a category error, not a rounding difference.",
      "present": true,
      "n_candidates": 9489,
      "admitted": 9408,
      "excluded": {
        "unparseable": 0,
        "noSigmaField": 81,
        "nonFiniteSigma": 0
      },
      "distribution": {
        "n": 9408,
        "min": 0,
        "max": 99,
        "mean": 1.5643,
        "quantiles": {
          "p05": 1.25,
          "p25": 1.37,
          "median": 1.54,
          "p75": 1.73,
          "p95": 1.73
        },
        "aboveZero": 9403,
        "aboveZeroPct": 99.95,
        "positivesOnly": {
          "n": 9403,
          "median": 1.54,
          "max": 99
        }
      },
      "digests": {
        "readingsSha": "ca5b4d6a32108f7ce2c0be5dd9d723779201e7903b4826d1a6fdfac8bb34234b",
        "bytesSha": "a3bf135e61f5682f6c6285f7a088c831a6fde2c522325858f482f885f3eb0511"
      },
      "unitOfReplication": {
        "clusterKeyAvailable": false,
        "reason": "a lens receipt records no reef/lens/corpus sha and no commit sha, so neither a condition cluster nor a de-duplication key can be derived from this surface",
        "remedy": "emit reefSha/libSha/corpusSha into the receipt at write time in prompt-lens.mjs — until then this corpus cannot carry a cluster-level rate",
        "n_raw": 9408,
        "n_distinct": null,
        "claimFence": "no cluster-level rate is computable here, so no independent-n claim may be made from this corpus"
      }
    },
    "measureHistory": {
      "id": "drift-sigma-ledger",
      "sourceGlob": "data/pmu/measure-history{,.archive,-archive}.ndjson",
      "sourceFiles": [
        "data/pmu/measure-history.ndjson",
        "data/pmu/measure-history.archive.ndjson",
        "data/pmu/measure-history-archive.ndjson"
      ],
      "sigmaField": "sigmaDrift",
      "producedBy": "the per-run measure ledger — appended by scripts/pmu/ledger-attest.mjs appendPricedRow() on the commit/fresh-pair path",
      "filterPredicate": "every ndjson row whose sigmaDrift parses as a finite number; rows flagged docsOnly / ingestSuspect / walkTruncated are ADMITTED-BUT-SEPARATED and reported, never silently pooled",
      "present": true,
      "perFile": {
        "data/pmu/measure-history.ndjson": {
          "present": true,
          "lines": 200,
          "admitted": 200
        },
        "data/pmu/measure-history.archive.ndjson": {
          "present": true,
          "lines": 989,
          "admitted": 989
        },
        "data/pmu/measure-history-archive.ndjson": {
          "present": true,
          "lines": 4887,
          "admitted": 4887
        }
      },
      "n_candidates": 6076,
      "admitted": 6076,
      "excluded": {
        "unparseableLines": 0,
        "nonFiniteSigma": 0,
        "missingFiles": []
      },
      "flagged": {
        "docsOnly": 3952,
        "ingestSuspect": 1072,
        "walkTruncated": 436
      },
      "distribution": {
        "n": 6076,
        "min": -7.81,
        "max": 64.23,
        "mean": 3.3228,
        "quantiles": {
          "p05": -2.33,
          "p25": -0.48,
          "median": 0,
          "p75": 2.2025,
          "p95": 25.78
        },
        "aboveZero": 2573,
        "aboveZeroPct": 42.35,
        "positivesOnly": {
          "n": 2573,
          "median": 3.24,
          "max": 64.23
        }
      },
      "distributionCleanOnly": {
        "note": "rows with none of docsOnly / ingestSuspect / walkTruncated set — the narrower read",
        "n": 1894,
        "min": -7.81,
        "max": 48.5,
        "mean": 1.2569,
        "quantiles": {
          "p05": -2.51,
          "p25": -1.2075,
          "median": -0.01,
          "p75": 1.68,
          "p95": 8.1705
        },
        "aboveZero": 926,
        "aboveZeroPct": 48.89,
        "positivesOnly": {
          "n": 926,
          "median": 1.745,
          "max": 48.5
        }
      },
      "digests": {
        "readingsSha": "644f35c858fba23c520a47225bb824af45e75ea1bbeea93d755f0da99cb9dec1",
        "bytesSha": "4cfecc2f16ed0a90600158a2631d107d5c937968f053f9538dac0092fc6149ed"
      },
      "unitOfReplication": {
        "clusterKeyAvailable": true,
        "tiers": {
          "apparatus": {
            "key": "(reefSha, libSha) — the setup held fixed across a long run",
            "rowsCarryingKey": 6076,
            "n_clusters": 2,
            "clustersAboveZero": 0,
            "clusterLevelAboveZeroPct": 0,
            "clusterMedianDistribution": {
              "n": 2,
              "min": 0,
              "max": 0,
              "mean": 0,
              "quantiles": {
                "p05": 0,
                "p25": 0,
                "median": 0,
                "p75": 0,
                "p95": 0
              },
              "aboveZero": 0,
              "aboveZeroPct": 0,
              "positivesOnly": {
                "n": 0
              }
            },
            "reading": "this is the count of genuinely distinct experimental configurations in the ledger. It is small. Any credibility claim phrased as \"N readings\" has to be read against it, because N readings spread over this many configurations is not N independent pieces of evidence."
          },
          "condition": {
            "key": "(reefSha, libSha, corpusSha) — the ledger's own full key",
            "rowsCarryingKey": 982,
            "rowsMissingKey": 5094,
            "missingKeyReason": "the older archive rows predate the corpusSha field (reefSha and libSha are present on every row); they are reported, not dropped, and they cannot enter this tier",
            "n_clusters": 610,
            "clustersAboveZero": 310,
            "clusterLevelAboveZeroPct": 50.82,
            "clusterMedianDistribution": {
              "n": 610,
              "min": -3.35,
              "max": 46.47,
              "mean": 3.9134,
              "quantiles": {
                "p05": -1.7365,
                "p25": 0,
                "median": 0.06,
                "p75": 3.39,
                "p95": 26.484
              },
              "aboveZero": 310,
              "aboveZeroPct": 50.82,
              "positivesOnly": {
                "n": 310,
                "median": 3.16,
                "max": 46.47
              }
            },
            "caveat": "corpusSha is the reality corpus OF THE COMMIT BEING MEASURED, so it changes with the commit by construction. This tier therefore collapses re-runs but lands at roughly one cluster per commit, and it does NOT absorb the autocorrelation between consecutive commits that share most of their reality corpus. Do not present it as if it did."
          },
          "distinctCommit": {
            "key": "commit sha — collapses repeat measurements of one recomputable computation",
            "n_clusters": 3055,
            "clustersAboveZero": 1055,
            "clusterLevelAboveZeroPct": 34.53,
            "clusterMedianDistribution": {
              "n": 3055,
              "min": -5.26,
              "max": 51.36,
              "mean": 2.2077,
              "quantiles": {
                "p05": -2.31,
                "p25": -0.31,
                "median": 0,
                "p75": 0.985,
                "p95": 20.765
              },
              "aboveZero": 1055,
              "aboveZeroPct": 34.53,
              "positivesOnly": {
                "n": 1055,
                "median": 2.74,
                "max": 51.36
              }
            },
            "repeatMeasurementRows": 3021
          }
        },
        "n_raw": 6076,
        "n_distinct_apparatus": 2,
        "n_distinct_condition": 610,
        "n_distinct_commit": 3055,
        "claimFence": "a rate from this corpus is publishable at CLUSTER level with n_raw stated beside n_distinct at the tier used, and the tier named. A pooled row-level percentage over the whole ledger pools repeat measurements of the same commit under a held-fixed reef and snippet library, which pmu-study-harness.mjs:28 names as pseudoreplication.",
        "residualAutocorrelation": "none of these tiers removes the fact that consecutive commits share most of their reality corpus and that docs-only commits recur. The tiers bound the overstatement; they do not license treating cluster-level observations as fully independent either."
      }
    }
  },
  "commitPagesJoinedToLedger": {
    "available": true,
    "method": "commit-page directory name (short sha) joined to the ledger row carrying the same short sha, taking that row's (reefSha, libSha, corpusSha)",
    "n_pages": 3554,
    "n_joined": 151,
    "n_clusters": 151,
    "clustersAboveZero": 104,
    "clusterLevelAboveZeroPct": 68.87,
    "joinedDistribution": {
      "n": 151,
      "min": -3.29,
      "max": 46.06,
      "mean": 5.156,
      "quantiles": {
        "p05": -2.24,
        "p25": -0.035,
        "median": 1.28,
        "p75": 5.065,
        "p95": 28.555
      },
      "aboveZero": 104,
      "aboveZeroPct": 68.87,
      "positivesOnly": {
        "n": 104,
        "median": 3.18,
        "max": 46.06
      }
    },
    "claimFence": "only the joined subset carries a cluster key; the unjoined pages remain a pooled read and must be reported as such"
  },
  "knownTruncation": {
    "file": "data/pmu/measure-history.ndjson",
    "cap": 200,
    "capSetAt": "scripts/pmu/ledger-attest.mjs appendPricedRow({ cap = 200 })",
    "restatedAt": "scripts/pmu/sigma-legend.mjs:39 (comment) above the MEASURE_HISTORY export at line 42",
    "effect": "a long-run distribution read from MEASURE_HISTORY alone sees only the most recent 200 rows and silently drops the rest",
    "observedHere": {
      "rowsInWindow": 200,
      "rowsTotalAcrossGlob": 6076
    },
    "remedy": "read the full glob data/pmu/measure-history{,.archive,-archive}.ndjson — the two archive files are DIFFERENT files (dot vs hyphen) and both are required for the full series"
  },
  "unitOfReplicationRule": {
    "source": "scripts/pmu/pmu-study-harness.mjs:28",
    "statedAs": "the unit of replication is the DISTINCT TILE, never a re-run of a recomputable computation (that is pseudoreplication); p-values come from the effect across distinct items",
    "appliedHere": "cluster by (reefSha, libSha, corpusSha); fall back to distinct commit sha; report n_raw beside n_distinct; compute the rate at cluster level; publish admitted vs excluded the way pmu-study-harness.mjs:384-388 does",
    "whyItBinds": "consecutive commits share most of their reality corpus, the reef and the snippet library are held fixed across long runs, docs-only commits recur, and the same commit sha is measured repeatedly — so a row-level rate pools highly autocorrelated observations and overstates how much independent evidence exists"
  },
  "headlineUnderAudit": {
    "asCirculated": "1203 readings · 84% above zero · median 3.51 · max 169",
    "resolvesTo": null,
    "finding": "no corpus in this repo reproduces that quadruple. Each recomputed distribution is recorded above with its own n, quantiles and digest; none matches on n, on the above-zero rate, on the median or on the max.",
    "status": "RETIRED PENDING RE-DERIVATION — the headline may not be republished until it is re-derived from a named corpus id in this record, at cluster level, with n_raw and n_distinct both stated.",
    "DECISION_POINT": {
      "id": "DP-SN1-1",
      "blocked": "which corpus the outward-facing credibility claim should quote",
      "why": "the three corpora answer different questions, and the choice is a positioning decision (what the σ is evidence OF), not a computation. Only the operator makes it.",
      "options_not_a_menu": "the computable recommendation is the drift-σ ledger at CLUSTER level, because it is the only corpus carrying a cluster key today; the other two require an emitter change first (see remedy fields).",
      "alsoBlocked": "whether prompt-lens.mjs should start writing reefSha/libSha/corpusSha into lens receipts so corpus B becomes cluster-able — an emitter change, out of scope for this audit."
    }
  },
  "elapsedMs": 9133
}
