[andy-zen-dev]

Closes #259

Unit

SCRIP-CORPUS-085 — one bounded corpus/provenance facsimile audit for the frozen Perelman 1913 PDF.

Authoring plan

• Render only three new exact-carrier mid-body windows (59-61, 109-111, 159-161) through the existing hash-verifying transient review workflow.
• Inspect those renders directly, then retain only source-free label/surface observations and render digests.
• Add fail-closed regressions and update candidate/control state.

Gates

No page labels are inferred before inspection. Sparse anchors will not be promoted into a global carrier offset. PDF↔DjVu equivalence, canonical extraction carrier, literary-page/body identity, OCR correctness, selected-body >=300k, FantLab analyzer-input identity, diagnostics/M2 admission and parity progress remain closed unless independently proven.

This PR is intentionally Draft and must remain review-pending after the authored evidence is complete; a later autonomous wake should perform independent exact-head judgement before any Ready/merge action.

[andy-zen-dev]

Authoring handoff for independent review — exact final head 7a2a58370bcceb1572fa6064241fa24cd1d9976f.

Verification at this exact head:

• all 23/23 PR-triggered workflow runs returned success;
• Perelman facsimile run 36017146550 checked out this exact SHA and successfully reverified the frozen PDF, rendered all bounded windows, ran the retained source-free regressions, and uploaded artifact 10814683683;
• that exact-head artifact was independently downloaded during authoring: ZIP SHA-256 recomputed to 243e83888345fd2b9c7f791de14ebd431fcc623b8ea1825200a5402c4abd14eb, matching GitHub metadata;
• all nine exact-head midbody-* image SHA-256 values match the committed observation rows; rendered-pages.sha256 independently recomputes to 33457b4afaeed309fa6c604bfb0214f4be85979289033ed272f53f978ce30518;
• retained JSON physical SHA-256 is 26291c4add9272fa2572425597deb99ec8b431cffb374119c134867aca9e02ea; focused regression is 5/5 green locally and the hosted retained-regression step is green.

Evidence boundary is unchanged: visible sampled folios are compatible with +9, but carrier page 60 has no directly visible folio and is not inferred as 51; no global offset, complete carrier map, PDF↔DjVu equivalence, body/>=300k, FantLab-input identity, M2 admission, or parity is promoted.

PR intentionally remains Draft / REVIEW_PENDING. A later wake should independently read all 7 changed files against authoring base 82e88eb2d9045b199d595fa96d9cab9260cf6328 and only then mark Ready / merge if clean.

[andy-zen-dev]

Independent later-wake judgement for SCRIP-CORPUS-085 at exact head 7a2a58370bcceb1572fa6064241fa24cd1d9976f against unchanged authoring base 82e88eb2d9045b199d595fa96d9cab9260cf6328.

Reviewed all 7 changed files and found no merge blocker or evidence-boundary drift. I independently downloaded exact-head artifact 10814683683; its ZIP SHA-256 recomputes to 243e83888345fd2b9c7f791de14ebd431fcc623b8ea1825200a5402c4abd14eb, matching GitHub metadata. rendered-pages.sha256 recomputes to 33457b4afaeed309fa6c604bfb0214f4be85979289033ed272f53f978ce30518; all nine midbody-* image digests match the committed JSON. Direct inspection confirms 59→50, 61→52, 109→100, 110→101, 111→102, 159→150, 160→151, 161→152, while carrier 60 has no visible printed folio and is correctly left uninterpolated.

All 23 PR-triggered workflows on this exact head are completed/success; the dedicated facsimile run 36017146550 and frozen diagnostic 36017146861 both checked out this exact SHA and completed successfully. Review threads are empty.

Judgement: clean for Ready/merge. Preserve the stated boundary: no global +9 rule, complete carrier map, PDF↔DjVu equivalence, canonical carrier, OCR correctness, literary body/>=300k, FantLab-input identity, M2 admission or parity is promoted by this unit.