Ratify pipeline-fidelity gate artifact-1 parity-replay outcome

[WilfordGrimley]

Description

Documentation-only PR ratifying the artifact-1 parity-replay outcome for the pipeline-fidelity gate (issue #154). The gate doc previously stated Artifact 1 as PENDING/methodology-only; the replay has since run and the owner has ruled on its outcome (issue #154 comments, 2026-07-22).

docs/pipeline-fidelity-gate.md — single source of truth for gate status/numbers per this repo's docs convention:

• §2 status table: Artifact 1 row PENDING → DONE, with the headline numbers (83.2% OCR-channel agreement on the 28,456-card reproducible subset; 373/41,586 (0.9%) unexplained divergences; 0/373 a wrong-printing vote).
• §2 gate-verdict paragraph rewritten to reflect current state.
• §3 (constants decision): all three MISSING constants are now decided — item 3 (deductive-backfill exclusion, landed via merged PR Add Stage D fallback channel calculator (pre-fire prep) #341) NOT restored; items 1–2 (RESOLUTION_FLOOR_DPI, EXCLUDED_RESOLVED_TAGS) ruled MUST-FIX (owner ruling 2026-07-22T23:47Z, sized against live data: 28 sub-floor cards / 47 custom-art+non-english cards / 75 union, all <0.05% of the 179,766-card eligible pool, zero live OCR votes affected) — fix is two one-line queryset excludes, in flight in a separate PR ahead of deploy.
• §4 rewritten from "open decision (b)" (methodology-only) into the full replay outcome: an explicit baseline-vs-method callout (pilot = legacy multi-channel OCR+phash engine; Stage D = OCR-only by design; this is a cross-method diff, not new-vs-itself), the divergence-bucket table, the owner's gate ruling text, both PR Fix OCR short-circuit over-skip and glued language-marker set code #340 root causes, and the code-fix-vs-live-cohort scope boundary.
• §7 (Chain): links to issue Pipeline-fidelity gate: dpi=250 parity replay + knowledge-inventory sweep (task #151) #154, PR Fix OCR short-circuit over-skip and glued language-marker set code #340, PR Add Stage D fallback channel calculator (pre-fire prep) #341.

docs/theory.md — added §7c, "Empirical parity-replay check (2026-07-22)": the replay as a durable technical finding — what it shows about the OCR-channel error term (ε₁, §7a) and the conservative-abstention property (g₄/g₅), plus the same explicit baseline-vs-method distinction (legacy multi-channel pilot vs. OCR-only Stage D). Numbers aren't duplicated — links to pipeline-fidelity-gate.md §4. Also appended one dated entry to the doc's pre-existing ## Status provenance log noting §7c, since theory.md is wiki-published (Theory).

Also touched, for consistency (one-liners, no restated numbers): CLAUDE.md, docs/README.md, docs/MANIFEST.md entries for pipeline-fidelity-gate.md.

Rebase note (2026-07-23): PR #341 merged to master after this PR was opened, touching the same file (its own landed version of the constant-#3 non-restoration rationale). Rebased onto current master; the one conflict (a redundant paragraph restating item-3's rationale, now superseded by #341's fuller landed inline version) was resolved by keeping master's landed wording and dropping the duplicate — no content lost on either side.

Facts verified directly against GitHub before writing (not taken on faith): gh pr view 340, gh issue view 154, gh pr view 341. All numbers match issue #154's own comments and the owner's in-session ruling verbatim.

Verification

• python3 .github/scripts/docs_lint.py — clean.
• python3 .github/scripts/tests/test_publish_wiki_link_rewrite.py — 4/4 passed.
• readme.md regenerate-and-diff parity — not applicable (no readme.md source-region touched), but ran python3 .github/scripts/publish_readme.py . + diff anyway: clean.
• Pre-commit hooks (prettier) reformatted the markdown tables on each commit; committed as-formatted.

Checklist

• I have installed pre-commit and installed the hooks with pre-commit install before creating any commits.
• I have updated any related tests for code I modified or added new tests where appropriate. (docs-only, n/a)
• I have manually tested my changes as follows: docs-lint + wiki link-rewrite parity + readme parity, all clean (see Verification above).
• I have updated any relevant documentation or created new documentation where appropriate.

Wiki

theory.md publishes to the wiki as Theory. Its own ## Status provenance section (the page's version marker) was updated in the same change — no separate wiki-page edit needed, it regenerates from source on merge.