Pass C deep metadata review — findings (2026-05-19)

Date: 2026-05-19 Reviewer: claude/cowork-pass-c-deep-review-2026-05-19 Scope: every metadata file under documents/, every paired document file. Structural audit ran across all 127 metadata files; deep textual review focused on the 12 verified files, plus spot-checks on needs_review files where prior discrepancy logs already flagged issues.

This pass picks up where the 2026-05-19 fidelity audit left off. The earlier audit identified four specific demotions (Flood, Finley, Popov, Garvey-no-change) and three filename/date/page-count discrepancies (Expos RICO, Selig-McCourt letter, Rose-Giamatti page count). This pass widens the lens to:

• automated structural checks across the full archive (hash drift, size drift, page-count drift, filename↔metadata date alignment, schema-required-field coverage);
• a deep textual review of every verified document against its underlying file;
• a systemic scan for the failure pattern that produced the four prior audit demotions ("status promoted without substantive content").

Headline result

No file hashes have drifted. All 127 metadata files with a recorded SHA256 match the on-disk file byte-for-byte. The archive's file-integrity rail is holding.

Substantive metadata accuracy is uneven. New findings span page-count errors (some non-trivial), one pre-existing discrepancy log that is itself inaccurate, three previously-undocumented historical/substantive errors in verified files, four verified files whose prose body still claims needs_review, and a systemic pattern of verified files containing explicit [deferred] placeholders in fields the schema requires to be substantive at that status level.

The headline action item: the verified-status bar needs to be re-applied to 4 of the 12 currently-verified files. None of them meets STANDARDS.md §6's required-content bar in its present form.

A. Automated structural checks (full archive)

Ran an audit script across every .md metadata file under documents/. Results:

Five new page-count mismatches surfaced. Two filename↔date mismatches are already logged in prior discrepancy notes and need no new action here.

B. New page-count errors (5)

All five files have correct SHA256 hashes — the underlying file is unchanged. The metadata's pages: value is simply wrong. None changes the substantive holdings recorded; all are factual record-keeping errors.

B.1 Popov v. Hayashi — meta says 11 pp, actual is 12 pp

documents/antitrust-and-courts/2002-12-18_caselaw_popov-v-hayashi.md

This is the most consequential of the five, because it intersects with an existing discrepancy log that is itself inaccurate. The 2026-05-19 fidelity audit's discrepancy note (2026-05-19_popov-fabricated-quotes-and-truncated-pdf.md) asserts:

• "The local PDF (...-statement-of-decision.pdf, 11 pages, ~95 KB) ends mid-discussion of Keron v. Cashman... without reaching the case's disposition section."
• "The local archive does not currently contain the case's actual disposition."

Both claims are demonstrably wrong against the file actually on disk:

• The PDF is 12 pages, not 11.
• The PDF reaches the full disposition on page 12, ending with: "the ball must be sold and the proceeds divided equally between the parties... The court retains jurisdiction to issue orders consistent with this decision. The ball is to remain in the custody of the court until further order. December 18, 2002 ____ Kevin M. McCarthy"

The SHA256 in the metadata (62d19ac7…ca00) matches the on-disk 12-page file exactly. The file has not been replaced; the discrepancy log was written without reading page 12.

The Westlaw additional file (...-westlaw.pdf) is 14 pages and likewise contains the full disposition verbatim on page 14.

Implications:

• file.pages should be 12, not 11.
• file.processing_notes claiming truncation should be removed.
• key_provisions[0] — currently a paraphrase with a disclaimer that "verbatim quote not yet verified — local PDF appears truncated" — can and should be replaced with the verbatim disposition text (which is present on PDF p. 12).
• The existing discrepancy note 2026-05-19_popov-fabricated-quotes-and-truncated-pdf.md should be amended: its core finding (that the prior key_provisions text was fabricated) stands; its secondary finding (that the file is truncated) is wrong and should be retracted.
• Re-promotion to verified becomes available without needing to acquire any new file — the verbatim disposition is already in the archive.

B.2 CBC v. MLBAM (8th Cir. 2007) — meta says 8 pp, actual is 13 pp

documents/broadcasting-and-ip/2007-10-16_caselaw_cbc-distribution-v-mlbam-8th-cir.md

The metadata says the affirmance is "in eight pages" (body) and pages: 8 (frontmatter). The actual PDF is 13 pages. The missing 5 pages are a dissent by Judge Colloton, beginning with "I respectfully dissent" structure and ending with "For these reasons, I would reverse the district court's grant of summary judgment in favor of CBC."

This is more than a page-count error: the metadata describes a unanimous panel affirmance ("Loken, C.J., Arnold and Colloton, Circuit Judges (Arnold, C.J., writing for the panel)") with no mention that Colloton dissented. The opinion is 2–1, not unanimous.

Implications:

• file.pages should be 13, not 8.
• The abstract, key_provisions, parties, and notes all need updating to acknowledge that Colloton dissented.
• The parties.panel description should say something like "Loken, C.J. (concurring), Arnold (writing for the panel) and Colloton (dissenting)" — or however the U.S. Reports / F.3d convention writes it. (Loken's role on the panel needs verification — he signed the majority based on the title line, but a quick read of the dissent suggests Loken was in the majority while Colloton dissented.)
• Per the smell test (STANDARDS.md §8), a metadata file that omits the existence of a dissent in a 2–1 appellate ruling cannot be cited as authoritative. Status should be demoted from verified to needs_review until the dissent is properly cataloged.

B.3 United States v. Cleveland Indians (SCOTUS, 2001) — meta says 21 pp, actual is 23 pp

documents/franchise-finance/2001-04-17_caselaw_united-states-v-cleveland-indians-scotus.md

U.S. Reports preliminary print runs pp. 200–222 (the case caption is at 532 U.S. 200 and the opinion ends at 222). That's 23 pages. Metadata says 21. The opinion file claim that the bench-slip majority is "~16 pp" is also off — actual is 18 pp.

Substantive content of the metadata is generally accurate — the holding text "Back wages are subject to FICA and FUTA taxes by reference to the year the wages are in fact paid" appears verbatim in the syllabus (p. 200) and at the end of the opinion (p. 220). But a smaller note:

• key_provisions[0] cites the holding at "532 U.S. at 208-209." The verbatim holding text doesn't appear at 208-209; it appears in the syllabus at 200 and at the conclusion of the opinion at 220. The page range 208-220 is what the syllabus itself cites for the discussion ("Pp. 208–220"). The pinpoint cite should be one of those, not 208-209.
• quoted_excerpts[0] and quoted_excerpts[2] cite "532 U.S. 200, at slip op. 1" — but the file being cited is the U.S. Reports preliminary print, not the slip opinion. Pinpoint citations should use U.S. Reports pages (likely p. 202 for the "FICA and FUTA impose excise taxes" passage; p. 200 syllabus for the holding).

Implications: page counts and three citation pinpoints need correction. No substantive demotion is required — the verbatim text is in the file, just mis-cited.

B.4 1989 SPBA Standard Player Contract — meta says 9 pp, actual is 11 pp

documents/contracts-and-rules/1989_contract_spba-standard-player.md

The file is a scanned PDF (Canon scanner output, no OCR layer — pdftotext returns no text). Metadata correctly notes nothing about substance pulled from the file, so the metadata's claims about the content of the contract are by definition not verified against this file. Page count should be 11.

This file should probably also have a processing_notes flag for the missing OCR layer (per STANDARDS.md §3.4: "For scanned PDFs without an embedded text layer, run OCR before storage (e.g., ocrmypdf) and store the OCR'd version. The fact that OCR was applied, and the tool used, must be noted in file.processing_notes."). Currently neither OCR has been run nor is the absence noted.

B.5 2008-03 MLB Constitution post-2005 amendment — meta says 20 pp, actual is 24 pp

documents/governance/2008-03_constitution_mlb-constitution-post-2005-amendment.md

PDF runs through Article XI (Miscellaneous) ending on what is labeled as p. 24. Metadata says 20 pp.

C. The 12 verified documents — deep textual review

Of the 12 currently-verified documents, 4 do not meet STANDARDS.md §6's required-content bar, 1 contains a substantive omission of dissenting opinion structure, and 5 are clean. The remaining 2 (Garvey, Cleveland Indians SCOTUS) have minor citation precision issues but are substantively correct.

C.1 Clean verifieds (5)

These were reviewed against their PDFs and the recorded quoted_excerpts and key_provisions were confirmed verbatim:

• 1922-05-29_caselaw_federal-baseball-club-v-national-league.md — all four quoted_excerpts verbatim at the cited pages; $80,000 verdict ✓; argued Apr 19, 1922 ✓; decided May 29, 1922 ✓; No. 204 ✓; "Error to the Court of Appeals of the District of Columbia (50 App. D.C. 165; 269 Fed. 681)" ✓; Holmes for unanimous Court ✓.
• 1953-11-09_caselaw_toolson-v-new-york-yankees.md — all four quoted_excerpts verbatim; argued Oct 13–14, 1953 ✓; decided Nov 9, 1953 ✓; Nos. 18, 23, 25 ✓; Burton dissent (joined by Reed) ✓; consolidated companion cases ✓.
• 1970-07-13_caselaw_salerno-v-american-league.md — all four quoted_excerpts verbatim including Friendly's famous "not one of Mr. Justice Holmes' happiest days" passage; panel (Waterman, Friendly, Hays) ✓; docket No. 818, 34653 ✓; argued May 26, decided July 13, 1970 ✓.
• 2003-02-10_caselaw_fox-sports-net-v-twins-8th-cir.md — both quoted_excerpts verbatim; panel (Wollman, Heaney, Melloy, with Heaney writing) ✓; submitted Dec 9, 2002, filed Feb 10, 2003 ✓; cross-appeals affirmed in full ✓.
• 2022-01-01_policy_joint-drug-prevention-treatment-program.md — preamble quoted_excerpt verbatim against PDF p. 1; "Effective January 1, 2022 through December 31, 2026" ✓; bit-identical MLBPA + MLB asset-server copies confirmed.

C.2 1882 Allegheny Base-Ball Club v. Bennett — two issues

documents/antitrust-and-courts/1882-11-18_caselaw_allegheny-base-ball-club-v-bennett.md

Issue 1: Stale prose claiming needs_review. Body line 136: "This document is needs_review. The transcription source (law.resource.org) is open and well-regarded, but the archive's standards (STANDARDS.md §2.4) require two independent sources for promotion to verified. A second source... should be obtained in a subsequent pass." But the frontmatter status is verified (since 2026-05-17), based on the SABR confirmation copy. The body text was not updated when the file was promoted.

Issue 2: Historical inaccuracy in interpretive commentary. Both notes (line 108) and the body (line 124) characterize the 1879 NL secret reserve agreement as an "eleven-man reservation." The 1879 NL secret reserve agreement was a five-player reservation per club (introduced at the September 1879 NL meeting in Buffalo). The reserve list was expanded later — to 11 players by 1883 with the Tripartite Agreement. Calling the 1879 introduction an "eleven-man reservation" conflates the 1879 origin with the 1883 expansion.

Independently, this entire interpretive aside about the reserve clause's introduction belongs in a separate document's metadata (or in research notes) per CLAUDE.md §1.8 — the Bennett metadata file should not be carrying editorial framing about the broader reserve-clause history.

Substantive accuracy on the document itself: the three quoted_excerpts are all verbatim from the law.resource.org PDF; case caption, court, judge, date, citation all match.

C.3 2001-05-14 MLBPA v. Garvey — one substantive omission

documents/arbitration-and-grievances/2001-05-14_caselaw_mlbpa-v-garvey-scotus.md

The metadata characterizes the decision as a clean per curiam reversal of the 9th Circuit. The Justia confirm copy shows the actual structure is per curiam + brief concurrence + Stevens dissent. The Stevens dissent reads in part: "I respectfully dissent... I find the Court's willingness to reverse a factbound determination of the Court of Appeals without engaging that court's reasoning a troubling departure from our normal practice."

Neither the concurrence nor the dissent appears in the metadata's parties, key_provisions, abstract, or notes. The opinion structure as recorded is incomplete.

Substantive accuracy of the recorded quoted_excerpt: verbatim against the Reporter's syllabus in the additional Justia copy ✓ (this was the question the existing 2026-05-19_garvey-file-organization-not-fabrication.md discrepancy resolved).

C.4 2007-10-16 CBC v. MLBAM (8th Cir.) — major omission

See §B.2. The metadata describes a unanimous affirmance; the actual ruling is 2–1 with Colloton dissenting. Pages 9–13 of the file are the dissent.

Recommendation: demote to needs_review until the dissent is properly cataloged.

C.5 2001-04-17 U.S. v. Cleveland Indians SCOTUS — minor citation issues

See §B.3. Page counts off; three pinpoint citations imprecise (citing "slip op." in a file that is the U.S. Reports preliminary print, and citing "532 U.S. at 208-209" for a holding that appears verbatim at 200 and 220). Verbatim text itself is present and accurate.

C.6 2002 Senate Judiciary Hearing 107-427 — verified but content empty

documents/legislation-and-hearings/2002_hearing_senate-judiciary-antitrust-mlb-107-427.md

Despite being marked verified, the metadata explicitly says:

• key_provisions[0]: "[Detailed content review of the 94-page hearing record deferred to future pass — typical structure: opening statements, witness testimony, prepared statements submitted for the record, appended documents.]"
• quoted_excerpts[0].text: "[Body text not yet transcribed.]"

STANDARDS.md §6 requires at verified: "key_provisions (≥1 entry)" with each provision "factual and attributable to a specific section of the document." A bracketed acknowledgment that the content was not reviewed is not such an entry.

Additionally: the metadata records date: "2002", date_precision: "year", with body saying "conducted in late 2001 / early 2002." The PDF's title page plainly shows February 13, 2002, single day. The metadata's date precision should be day. (And the filename 2002_hearing_... would benefit from being renamed 2002-02-13_hearing_... per NAMING.md §2.1 — preserving the old filename per the convention.)

Recommendation: demote to needs_review until at least one substantive key_provisions entry and one verbatim quoted_excerpts entry are populated. Date precision upgrade is a separate small fix.

C.7 1996-12-07 MLB CBA (1997–2001) — three issues

documents/cbas/1996-12-07_cba_mlb-cba-1997-2001.md

Issue 1: [deferred] placeholders in required content fields. key_provisions[5] says "Article VI — Salaries (TOC reference): minimum salary, salary arbitration, etc. [full provisions deferred to future review pass]." key_provisions[6] says "Article VII — Expenses and Expense Allowances [deferred]." key_provisions[7] says "Articles VIII-XXI [deferred — full structural mapping requires further content review]." Same STANDARDS.md §6 violation as C.6.

Issue 2: Stale needs_review prose (body line 148). Frontmatter is verified since 2026-05-17.

Issue 3: Factual slip in abstract. Abstract refers to "Pittsburgh Pirate Pete Rose Jr." — Pete Rose Jr. played for the Cincinnati Reds (a brief 1997 call-up) and the Chattanooga Lookouts in the Reds' minor league system; he never played for the Pirates. This is editorial scenery the metadata should not be carrying anyway (CLAUDE.md §1.8) — but as long as it's in there, it should be factually correct.

The actual quoted_excerpts (preamble, Article V) are verbatim against the PDF.

C.8 2002-09-30 MLB CBA (2002–2006) — three issues

documents/cbas/2002-09-30_cba_mlb-cba-2002-2006.md

Issue 1: [deferred] placeholder in required content field. key_provisions[12] says "[Articles XIII through XXVIII and Attachments — including the Joint Drug Prevention and Treatment Program — not yet itemized in this metadata pass; full structural mapping requires future content review.]" STANDARDS.md §6 violation.

Issue 2: Stale needs_review prose (body line 142).

Issue 3: Possibly-incorrect historical claim. The metadata's notes says this was "the first time since 1972 that MLB had avoided a labor stoppage in a CBA-expiration cycle." That depends on how one counts: the 1976 CBA followed a spring 1976 owner lockout; the 1990 CBA followed a 32-day spring 1990 owner lockout. The 1985 strike was a 2-day stoppage. The plainer formulation — that this was the first CBA reached without an in-cycle work stoppage in a long time — should replace the specific year claim, or the year claim needs sourcing. Either way: editorial framing of this kind is a CLAUDE.md §1.8 concern.

The single quoted_excerpt citation reads "2002-2006 Basic Agreement, preamble (per BoB description; text in archive PDF)" — soft framing for a verified excerpt. The quoted text should appear verbatim in the PDF or be removed. (I did not separately verify whether the preamble text quoted matches the PDF; the metadata's own equivocation invites that check.)

C.9 2022-03-10 MLB CBA (2022–2026) — two issues

documents/cbas/2022-03-10_cba_mlb-cba-2022-2026.md

Issue 1: Placeholder quoted_excerpt. quoted_excerpts[0].text says "[Body of the agreement; full text not transcribed in this metadata file — see PDF for verbatim provisions]" — explicitly not a verbatim quote. STANDARDS.md §6 expects verbatim text in quoted_excerpts (the field is optional in principle but the worked example in METADATA_SCHEMA.md §3 treats it as expected for verified case-law/document files).

Issue 2: file.pages: 0 for a 440-page agreement. PDF actually has 440 pages. Same minor issue affects JDPTP (pages: 0 for a 137-page program).

Issue 3 (minor): Stale needs_review prose (body line 165). Less severe than the others because the body's "Verification status" section explicitly explains the constraint (MLBPA-hosted copy + needing an MLB-side mirror) and the SABR bit-identical confirmation was retrieved subsequently. But the prose still reads as if the second source hasn't been found.

D. The systemic pattern

Of the 12 verified files, 4 contain [deferred] / not yet transcribed / not yet itemized placeholder text in key_provisions or quoted_excerpts — fields that STANDARDS.md §6 requires to be substantive at the verified status level:

• 1996-12-07 MLB CBA
• 2002-09-30 MLB CBA
• 2022-03-10 MLB CBA
• 2002 Senate Judiciary Hearing 107-427

A broader scan finds the same [deferred] / [not yet] / [full content review deferred] pattern in 35 needs_review files, which is less serious in absolute terms (the status itself disclaims completeness) but still adds up to ~30% of the archive carrying these placeholder fields. If the archive's eventual public-facing front-end renders key_provisions or quoted_excerpts directly, those placeholders will leak into the user-facing presentation.

This is the same failure pattern that produced the four 2026-05-19 fidelity-audit demotions (Flood, Finley, Popov, Garvey). The mechanism in each case was: status was promoted to verified on the strength of the source-confirmation work, before the substantive content fields were verbatim-confirmed against the document. The bar at verified requires both the source-confirmation work and the substantive content. Promoting on the first is what creates the audit risk.

Recommendation: Add a check item to the verified-promotion workflow: "key_provisions and quoted_excerpts contain no [deferred] / not yet / TBD bracketed text", and have this checked before status is promoted, not as a follow-on cleanup.

E. Stale "needs_review" prose in verified files

Four verified files still carry prose that explicitly claims the file is needs_review:

• documents/antitrust-and-courts/1882-11-18_caselaw_allegheny-base-ball-club-v-bennett.md (body line 136)
• documents/cbas/1996-12-07_cba_mlb-cba-1997-2001.md (body line 148)
• documents/cbas/2002-09-30_cba_mlb-cba-2002-2006.md (body line 142)
• documents/cbas/2022-03-10_cba_mlb-cba-2022-2026.md (body line 165)

The prose was written when the file was at needs_review and not updated when the file was promoted to verified. Mechanically the prose is in the document body, not in YAML-required fields, so the schema validates — but a reader reaching the bottom of any of these four files sees a contradiction between the frontmatter and the body.

F. Page-count drift summary

G. Recommended remediation

Demote to needs_review (because the file does not currently meet verified-status content requirements):

1. documents/cbas/1996-12-07_cba_mlb-cba-1997-2001.md — [deferred] placeholders in key_provisions[5]–[7].
2. documents/cbas/2002-09-30_cba_mlb-cba-2002-2006.md — [deferred] placeholder in key_provisions[12]; "per BoB description" softness on the quoted_excerpt.
3. documents/cbas/2022-03-10_cba_mlb-cba-2022-2026.md — placeholder quoted_excerpts[0].
4. documents/legislation-and-hearings/2002_hearing_senate-judiciary-antitrust-mlb-107-427.md — placeholder in both required substantive fields.
5. documents/broadcasting-and-ip/2007-10-16_caselaw_cbc-distribution-v-mlbam-8th-cir.md — material omission of a dissent on a 2–1 panel.

Correct in place (no demotion needed):

6. documents/antitrust-and-courts/2002-12-18_caselaw_popov-v-hayashi.md — fix pages: 11 → 12, drop the truncation processing_note, replace key_provisions[0] paraphrase with the verbatim disposition from PDF p. 12. (And amend the existing discrepancy log 2026-05-19_popov-fabricated-quotes-and-truncated-pdf.md to retract the truncation claim.)
7. documents/franchise-finance/2001-04-17_caselaw_united-states-v-cleveland-indians-scotus.md — fix page counts; revise pinpoint citations (slip op. → U.S. Reports pagination; 208-209 → 200 or 220 depending on which appearance is being cited).
8. documents/antitrust-and-courts/1882-11-18_caselaw_allegheny-base-ball-club-v-bennett.md — fix stale needs_review body text; correct the "eleven-man reservation" historical claim (or remove the editorial aside per CLAUDE.md §1.8).
9. documents/arbitration-and-grievances/2001-05-14_caselaw_mlbpa-v-garvey-scotus.md — add mention of the Stevens dissent and the brief concurrence to parties/abstract/notes.
10. documents/contracts-and-rules/1989_contract_spba-standard-player.md — fix pages: 9 → 11; add processing_notes flagging the missing OCR layer (per STANDARDS.md §3.4).
11. documents/governance/2008-03_constitution_mlb-constitution-post-2005-amendment.md — fix pages: 20 → 24.
12. documents/drug-and-conduct/2022-01-01_policy_joint-drug-prevention-treatment-program.md — fix pages: 0 → 137.

Body-prose-only fixes (verified status retained):

13. The four stale-prose files in §E. Body sections that say "needs_review" need to be updated to match the actual status.

H. What this pass did NOT cover

Scope honesty:

• Verbatim text matching for needs_review documents: not done at scale. I read multiple needs_review files but did not run the full document-PDF spot-check on every one. The CBC v. MLBAM dissent finding is an example of what such checks turn up; there are 114 needs_review files and on the pace of this pass, full textual review of all of them would be 2–3 more sessions.
• Historical fact-checking of every metadata-supplied historical claim (e.g., "first time since 1972", "Pittsburgh Pirate Pete Rose Jr.", "eleven-man reservation", "1879 secret reserve agreement"): not done at scale. The three flagged in this pass surfaced incidentally during deep review of specific files. The archive's metadata generally carries a lot of editorial scenery (CLAUDE.md §1.8 concern); a systematic pass that either removes or fact-checks each such claim is its own project.
• Cross-references in related_documents arrays: not checked. A pass that confirms every related-document slug points to an existing file would catch any broken cross-links.
• status_history append-only integrity: not checked. A pass that confirms each promotion/demotion is logged in PROVENANCE_LOG.md would close the audit-trail loop.

These are appropriate scopes for follow-on passes.

Cross-references

• Prior audit notes from the same day: 2026-05-19_flood-syllabus-elision.md, 2026-05-19_finley-case-number-and-panel.md, 2026-05-19_popov-fabricated-quotes-and-truncated-pdf.md (this pass partly retracts §B.1), 2026-05-19_garvey-file-organization-not-fabrication.md (this pass adds the Stevens-dissent omission).
• Backfilled audit notes from earlier passes: 2026-05-19_expos-rico-complaint-2000-vs-2002.md, 2026-05-19_selig-mccourt-letter-date.md, 2026-05-19_rose-giamatti-page-count.md.
• Governance: CLAUDE.md, STANDARDS.md (especially §1, §6, §8), METADATA_SCHEMA.md §2 and §3, TAXONOMY.md.