[amaybaum]

Draft. Serialized unit 4: #1028 unit 1 (orphan 35), authorized by the owner on 2026-11-02. It must not be merged without the owner's exact-SHA approval.

This PR implements the independently accepted correction proposal (r9) exactly. Before r9 was accepted, the proposal went through:

• the C2 gate and its review;
• reviews r4–r8 and a final-gate confirmation, with a fresh independent reviewer on r7 to r9;
• owner scope decisions at each round: Option A, Option 1, Option C, Option A for the exact-Markov sites, and Option A for the Medicine sites.

It is based on main c287a577. Every one of its 84 old spans was re-verified to occur exactly once at that commit before editing.

What the correction establishes in the manuscripts

• Fast mixing gives suppression, not exclusion. Main's erasure bound gives δ_k ≤ δ_0 e^{−(k−1)τ_S/τ_B} for orders k ≥ 2 only. No regime of it gives exact vanishing, so exact ¬C4 and M_t = 0 claims are replaced with suppression.
• (C2-slow) vs (C2-structural).
  • The observer exclusion holds only with (C2) read in its timescale form, (C2-slow). Its failure follows from the incompatible timescale orderings, under the stated mixing and rate assumptions, in the coarse-grained effective-channel description.
  • (C2-structural), persistent accessible memory, is left unresolved: fast mixing is not shown to remove it.
  • C4 failure is not established.
  • Scope the past-hypothesis replacement to its observer-selection role #1024's observer-selection result stays conditional.
• P-divisibility, using only Main:455–459, ‖T^(k) − T^k‖_TV ≤ (k−1)ε under its stated hypotheses.
  • ε = 0, the exact reset, is the specialisation of that theorem with the same T, π_H and hypotheses. It gives the Markov relation and P-divisibility.
  • For ε > 0 the bound limits the departure from the benchmark but does not determine P-divisibility. No approximate-P-divisibility claim is made.
• No sharp thresholds from τ_B alone.
  • Spin-chain (ch15 and Complexity): chain lengths are labelled as illustrative estimates.
  • Medicine and ch16: cell-identity behaviour is stated as model-dependent unless a mechanism establishes more.
  • ch11:148: record absence is an explicit premise.

Files (34)

• Manuscripts: 18 .md files, 86 line edits in total, including 26 FULL mirrors.
  • Papers: Main 11, Explainer 9, Complexity 6, Substratum 3, Structure 3, Medicine 2.
  • Book chapters and FULL.
• LEDGER.json: three fingerprints move, computed by proof_census.py: Main:165 → 78863869e1258fa3, Main:756 → c6f15bd32bf1f7ae, Main:758 → a2f5c0c423aa9ca2. All 134 rows equal the census.
• sc42 result: predicates 280 → 303. check_sc42_integration.py passes 641 checks.
• Rebuilt .tex/.pdf: Main, Complexity, Substratum, Explainer, Structure, Medicine and FULL. Canonical toolchain, 0 dropped glyphs, none provisional.

Local checks (writer):

• architecture checker: 0 violations;
• toolchain, staleness, mirror, voice, claims, duplicate, coverage, Lean-manuscript census, artifact placement and dependency-label checks: OK;
• added-text register scan: 0 hits;
• all 488 rows of the reference classification match the worktree: 417 supported, 65 conditional, 6 deferred, 0 contradictory.

legacy_records_check and citation_roles_check need full history, so CI decides them.

Not in this PR (still tracked on #1028):

• GR:20 and Main:218;
• the other orphans;
• deferred items ch11:146, Complexity:278, README:90, ch03:60, ch13:106, Medicine:35 and ch16:184.

Explainer:1019 is corrected here by owner decision, with the cross-reference recorded on #1029.

Before the owner's decision: independent exact-head review, and CI including Lean. The five holds (#1009, #1023, #1011, #1003 and #918) are unchanged.

---

Generated by Claude Code

[amaybaum]

Serialized unit 4 checkpoint (2026-11-02): head 77edc1a1004aa7312bd3ec24eefcaa5a1e6db8af. Awaiting the owner's exact-SHA merge decision; not merged.

Independent exact-head review: READY, no nits.

• Manuscript text. Applying the accepted r9 proposal's 84 old→new hunks to c287a577 reproduces all 18 changed .md files byte for byte.
• Generated files.
  • Fresh pandoc runs reproduce all 7 .tex files, including their source-sha lines.
  • LEDGER changes exactly the three stated rows. Their values equal proof_census.py, and coverage_check passes.
  • The sc42 checker reproduces the committed result JSON, which changes only by its 23 added predicates (280 → 303).
• Checks. The architecture checker reports 0 violations, and the diff contains no unexpected file.
• Science. The owner's distinctions hold at every spot-checked site: Main:76, :171, :451, :453, :756 and :758; ch15:111–116; Medicine:307 and :483; ch11:148.
  • C2-slow is conditional, and C2-structural is unresolved.
  • C4's failure is not established.
  • ε = 0 is stated as the specialisation of Main:455–459; for ε > 0, P-divisibility is left undetermined.
  • No sharp threshold is inferred from τ_B alone.
• Mergeability. Main is still c287a577, and the merge is clean.

CI. Run 41412998179 (pull_request, verify) passed: 25 jobs succeeded and 1 was skipped as expected, including the Lean kernel and Mathlib bridge. GitHub reports the PR as mergeable, with state clean.

Reference classification. The occurrence-level sweep of 488 rows (417 supported, 65 conditional, 6 deferred, 0 contradictory) and the deferred follow-ups, each with its accountable issue, are preserved on #1028 in comments 6572579040 and 6572625127.

The five holds are unchanged, and #1028 stays open for its remaining obligations.

---

Generated by Claude Code

[amaybaum]

Merge record. Merged at the owner-approved exact SHA 77edc1a1004aa7312bd3ec24eefcaa5a1e6db8af as a normal merge commit. Immediately before the merge, the head still matched the approved SHA and all 26 check runs were green (25 succeeded, 1 skipped as expected).

• Merge commit: 01a096a95a02caa1c471d892f34db919593783d9.
  • Parents: c287a577 (main) and 77edc1a1 (the approved head).
  • Its tree is identical to the approved head's tree.
  • It brings the 34 planned files into main.
• Post-merge CI: push CI run 41423964209 on 01a096a9 passed, with 25 jobs succeeding and 1 skipped as expected, including the Lean kernel and the Mathlib bridge.

The reference classification stays preserved on #1028 (comments 6572579040 and 6572625127). #1028 stays open for its remaining obligations.

---

Generated by Claude Code