Run 02: recover the editions before revising them

Date: 23 September 2026. Scope: archival recovery, not a new proof verdict.

Why this run comes first

The four papers had already changed before Run 01 pinned their current files. That baseline prevents further drift but does not explain what changed since first publication. This run recovers the earlier artifacts and makes them available without changing the current manuscripts.

Reproducible sources

• GitLab history: insight-forge/boxlab/fprd-lab at a5ceeaedc42bb1bc5c91e3d550a6b4d85703a3d6.
• Site history through the published review baseline: 45fffe26aabd79d67d2f8962575005182ec34c6d.
• Extractor: scripts/archive-scaffold-revisions.py in the Site source.
• Archive: history/manifest.json, with full SHA-256 hashes, original paths, Git blob identities, commit identities/dates, and content-addressed files.

The extractor scans the four publication directories for PDFs, main.tex, and revision logs. It also preserves per-paper imported/expanded HTML bodies, afterwords, the direct paper’s maintained HTML source, and legacy EPAL render sources. It records content changes, not every unrelated repository commit. JSON-body entries are extracted UTF-8 strings; their source blob identifies the containing JSON file, not a standalone Git blob of the fragment. Resolved article snapshots combine the active imported/expanded body with the afterword actually present at that source revision, including removal of an afterword. Their component-blob references identify those inputs. These snapshots still exclude the surrounding Site layout and are not HTTP captures.

Recovered lineage and documented changes

Real-time multitape transfer

The source history begins at 4615154 on July 30. July 31 revisions cover tape-level reconciliation (4cefe28) and the T90 composition (0ac821f). August 5 includes rewritten exposition, bibliography, source links, and the synchronized formalization release (14dc9d9, c340fbc, 6376bbd). These revisions reuse the established PDF filename. The early Site route linked to GitLab’s mutable main URL, so the current link cannot identify the bytes seen by an earlier reader.

Direct palindrome scaffolding

The compact first release (a287a61, August 21) and expanded specialist release (687ce12, also August 21) precede the separately named Draft 2 (2e0e31f, August 22). The later preserved Draft 1 is not a substitute for both earlier byte sequences. Both are recovered from Git objects.

The repository’s Draft 2 log explicitly records a receipt-closure repair (the prior use of bridge shadow lacked its longest-suffix hypothesis) and the addition of typed transition receipts to avoid unbounded suffix-link traversal. It also records changes to empty records, opacity, and the radius inventory. These are historical assertions to trace, not newly certified proofs in this run.

Later Site editions include the virtual-initialization folio (Draft 3), an applicability folio (Draft 4), intermediate HTML-only interface/certificate changes, and the maintained-source Draft 5. The archive retains the actual artifacts rather than generating new PDFs for intermediate HTML revisions.

Persistent scaffold causal algebra

The first source release is 4d08d0d (August 21), followed by Draft 2 at 2e0e31f and Draft 3 at f69e26f (August 23). The Draft 2 log explicitly removes the companion-dependent palindrome corollary and treats that application conditionally. Draft 3 corrects Proposition 8.5’s finite carrier coordinates. The later Site corrected Draft 4 adds an application-scope folio; its body and the expanded HTML need to be compared statement by statement rather than assumed identical.

Unary scaffold coherence

Draft 1 starts at f69e26f (August 23). Its later expanded HTML and appended PDF afterword are separate changes even though the draft label remains 1. The review must track inherited causal-observation assumptions, not infer that a shared draft label establishes unchanged content.

Shared Site changes

• The August 24 library import (c38b355) is not the first publication of these papers. It imported already revised editions.
• The expanded reading-edition commit (473e2ff) and its September 5 edition label are different date evidence; neither should silently replace the other’s date.
• The September 9 afterword commit (a092c34) changed the four PDFs under unchanged filenames. Both pre-afterword and post-afterword bytes are now preserved.
• Source-only attempts are not automatically public releases. In particular, the abandoned T160/T161 edit attempt described in Run 01 was not deployed and is not a new edition of these four papers.

How previous revisions will be handled

1. Preserve original bytes and original draft labels. Identify same-label or same-filename variants by source date, commit, and full content hash.
2. Preserve HTML-only changes as HTML fragments or render sources. Do not manufacture a historic PDF and imply that it was circulated.
3. Build a cumulative statement-level ledger: old statement, new statement, reason, affected dependencies, and whether the change is mathematical, explanatory, or formatting-only. Existing logs seed this ledger but do not replace comparison of actual versions.
4. Use that ledger for concise end-of-paper histories in the next numbered, matching PDF/HTML editions. Do not reset the history at today’s baseline.
5. Continue reviewing all four proofs together. A change in the direct route does not by itself alter the classical-recognizer transfer route.

Limits and next obligation

This recovers repository artifacts, not exact reviewer receipt histories. Commit dates do not prove when a website was deployed. Historical HTML fragments do not preserve every wrapper, stylesheet, or exact HTTP response. The statement-level classification and cross-paper mathematical impact analysis are unfinished. H01 therefore remains active, ahead of R02.

The existing preservation test checks that all four current article bodies and linked PDFs still match Run 01. Archive tests check every content hash and ensure each recorded event resolves to a preserved artifact.