# Independent factual validation (summary)

Run on the exact data files of this release (SHA-256 of the main file `e81a8510ddf4f45b0f84d87e1f2135e79637ded19a3f91bcafd0c7ecbb1bc364`).

**Method (fixed before the first fetch; one clarification of the INFERRED check is recorded in the validation record).** Rows were drawn by SHA-256 order of `row_id`, not by outcome: 30 tier-B rows (6 from each of five eras between 2004 and 2025) and 10 tier-B2_FTD rows. For each, the first and last cited SEC document was downloaded fresh and checked: the issuer CIK must appear in the document and, for tier B, the ticker must appear as a whole word (for B2_FTD the first company-name token). Separately, 20 INFERRED rows were checked structurally: each must sit in a gap of 366 days or less between two A/B rows with the same CIK and ticker.

**Result.** 40 of 40 sampled rows verified, 0 partial, 0 contradicted, 0 not retrievable. INFERRED structural check: 20 of 20.

**Rows whose evidence references were resolved.** 152 rows (provisional research windows and rows whose references were stored as a document hash or an internal file path) had their SEC document URLs resolved from Persistfolio's existing research and were all checked (full population). For 94 hash-resolved rows, the freshly downloaded SEC document hash equals the hash recorded in the research file in 94 cases. Pre-registered text test (issuer CIK printed in the document and the ticker or name token present): 123 verified, 29 failed, 0 row has no recoverable URL (listed in EXCEPTIONS.csv). The 29 failures are documents (old .htm filing bodies) that do not print the CIK in their text; a weaker secondary test (CIK is the URL folder and the ticker is in the text) passes for 29 of 29. The pre-registered verdict for this check is REVIEW; the secondary test is reported as a weaker supplement, not a replacement.

**What this does and does not show.** It shows the cited filings contain the claimed CIK and ticker for the sampled rows. It does not show exchange listing, that the symbol traded on every day of the segment, that a share class is correct, or accuracy of unsampled rows. The sample is small; no error rate for the full file is claimed. Tier C rows (unresolved) were not tested because they assert nothing. The method looks at one evidence family; it is not a third-party audit.

# Quality report

## Coverage (member-days, 2004-09-30 to 2025-09-30; 3,851,484 days, 978 securities)
| Cumulative | Days | Share |
|---|---|---|
| A + B (strict) | 2,817,317 | 73.15% |
| + B2_FTD | 3,333,020 | 86.54% |
| + INFERRED | 3,651,752 | 94.81% |
| C_UNRESOLVED | 199,732 | 5.19% |

By tier: A 9; B 2,817,308; B2_FTD 515,703; INFERRED 318,732; C_UNRESOLVED 199,732.
Rows: 36,953; rows with a ticker: 33,135; distinct tickers 997; distinct issuer CIKs 964.
Securities with no identity evidence on any day: 13.

## By calendar year
| Year | Member-days | A+B | B2_FTD | INFERRED | C | % A+B | % unresolved |
|---|---|---|---|---|---|---|---|
| 2004 | 46,468 | 32,826 | 0 | 8,423 | 5,219 | 70.642 | 11.231 |
| 2005 | 182,228 | 145,215 | 6,934 | 18,445 | 11,634 | 79.689 | 6.384 |
| 2006 | 182,500 | 141,018 | 12,791 | 15,550 | 13,141 | 77.270 | 7.201 |
| 2007 | 182,500 | 140,463 | 17,025 | 13,473 | 11,539 | 76.966 | 6.323 |
| 2008 | 182,995 | 134,870 | 21,446 | 16,370 | 10,309 | 73.701 | 5.633 |
| 2009 | 182,428 | 130,475 | 24,413 | 16,001 | 11,539 | 71.521 | 6.325 |
| 2010 | 182,319 | 128,533 | 29,747 | 15,689 | 8,350 | 70.499 | 4.580 |
| 2011 | 182,500 | 129,117 | 30,390 | 14,820 | 8,173 | 70.749 | 4.478 |
| 2012 | 183,012 | 132,302 | 28,036 | 13,083 | 9,591 | 72.291 | 5.241 |
| 2013 | 182,500 | 129,900 | 30,894 | 13,806 | 7,900 | 71.178 | 4.329 |
| 2014 | 182,921 | 128,190 | 31,750 | 14,756 | 8,225 | 70.079 | 4.496 |
| 2015 | 183,520 | 128,801 | 30,414 | 15,990 | 8,315 | 70.184 | 4.531 |
| 2016 | 184,558 | 132,599 | 25,818 | 14,970 | 11,171 | 71.847 | 6.053 |
| 2017 | 184,326 | 124,265 | 34,155 | 14,188 | 11,718 | 67.416 | 6.357 |
| 2018 | 184,326 | 125,873 | 34,744 | 14,504 | 9,205 | 68.288 | 4.994 |
| 2019 | 184,329 | 128,025 | 31,001 | 14,336 | 10,967 | 69.455 | 5.950 |
| 2020 | 184,839 | 143,144 | 17,082 | 15,865 | 8,748 | 77.443 | 4.733 |
| 2021 | 184,326 | 137,916 | 21,826 | 14,355 | 10,229 | 74.822 | 5.549 |
| 2022 | 183,872 | 138,786 | 22,751 | 14,066 | 8,269 | 75.480 | 4.497 |
| 2023 | 183,597 | 139,649 | 21,195 | 14,896 | 7,857 | 76.063 | 4.279 |
| 2024 | 184,101 | 141,196 | 19,432 | 18,063 | 5,410 | 76.695 | 2.939 |
| 2025 | 137,319 | 104,154 | 23,859 | 7,083 | 2,223 | 75.848 | 1.619 |
| ALL | 3,851,484 | 2,817,317 | 515,703 | 318,732 | 199,732 | 73.149 | 5.186 |

2004 begins on 2004-09-30.

## Checks run on this release (see certification record)
Exact hash binding of every file; security-id set equals Cohort 1.1 (978 of 978); per-security segment union equals Cohort membership days (no gap, no overlap); tier days reconcile to the frozen build manifest; row_id unique; ticker/CIK blank exactly on unresolved rows; no CUSIP column, token or word; no segment dated in 2026; deterministic double rebuild.

## Known limitations (read with the tier table)
- Identity is SEC-evidenced, not exchange-certified. A Form 4 issuer symbol shows the symbol an insider filing used.
- 25,251 rows have `UNCLASSIFIED_COMMON`: the issuer's common stock, class not established. Multi-class issuers need your own check.
- INFERRED rows are unwitnessed gaps between two SEC-evidenced anchors with the same issuer CIK and ticker. They are labelled research evidence: the identity is carried across the gap, not witnessed in it. Persistfolio's own research notes advise against using this tier for certified trading, price joins or corporate-action rights; treat it the same way. Corporate actions, symbol or class changes inside a gap were not excluded; in Persistfolio's 2004-2005 pilot screens of bracketed gaps, 20 of 25 held SEC 8-K text with corporate-action keywords (noisy) and 4 of 10 held high-sensitivity 8-K items (2.01, 3.01 or 5.03) needing review. That is a signal of residual risk, not an error rate for this release.
- 327 tier-B rows (95,823 days) are provisional research windows (see `evidence_statement`): SEC filings show the identity at points inside the window, and the window's bounds were set by Persistfolio's research and not verified day by day. They are 2.49% of all member days and 3.4% of tier-B days, and they are included in the tier-B total in the table above; subtract them if you want only directly witnessed windows. They are the weakest part of tier B.
- Every non-INFERRED row carries at least one SEC document URL.
- Third-party web pages appear in `supplementary_references` on 50 rows; they are context only, not tier evidence, and their content is not redistributed.
- Right edge: coverage ends 2025-09-30 by design.
- No independent third-party audit of the full dataset. Independent factual spot-checks are summarized in `docs/VALIDATION_SUMMARY.md` (small samples; they do not establish accuracy for the whole file).