[amaybaum]

Guiding draft PR for the ARROW-1 manuscript correction round, #1112.

Status: WIP / UNVERIFIED. The owner's follow-up rulings (#1112 c.6650301410) and the C1 ruling (#1112 c.6651202921) are applied. The exact-head review of a299ddf4 is recorded below. No merge is authorized.

• Head: a299ddf49cd668ceafe22b9e769a4714717e9ed3
• Base: main at f5c281ae10a913ec68fa41b3b839ad0b164979cd. main has since moved to 53a3eb3d (#918 B1: retain the [CE]/[CR] evidence commits by ancestry only (history-only merge, no file changes) #1117, a history-only merge whose tree is unchanged). The branch needs an ordinary merge of main, with a delta check, before a merge gate.

Authority

• Plan: #1112 c.6643840329
• Option 3: c.6649271094
• Phase 2 rulings: c.6649643759
• Follow-up rulings: c.6650301410
• C1 ruling: #1112 c.6651202921
• Reviews:
  • Recovery note: c.6649600806 and c.6649698455.
  • Phase 2 at bd926c76: scientific c.6650236439, minimum-change c.6650241392.
  • Follow-up at 98d04808: c.6650406439 (blocking item C1).
  • Exact head a299ddf4: independent scientific review ACCEPT, integration review ACCEPT WITH CHANGES (PR metadata only, applied here). Recorded in the comments.
• Writer: writer-arrow

Commits

Phase 1: b398082d, bd398f05, 37b39d1c.

Phase 2:

• e133ffbc: record.
• d3e13c99: Main and LEDGER.
• eb74b3e7: papers.
• 4081cba0: book and FULL.
• e4ab279a: rebuild.
• 24452b01: CHANGES.md.
• 6010cf44: audit-census counts (ruling A4).
• bd926c76: FLRW spelled out.

Follow-up:

• f6f7dd13: rulings A1, A2, A5, B1–B3, B5 and N1–N8, mirrored to FULL, with Substratum, SM, Methodology, Explainer and the book rebuilt.
• a762d418: SC-42 result regenerated (A3), as a separate commit.
• 98d04808: CHANGES.md updated.

C1 ruling:

• ca21185e: the ch18:214 / FULL:5027 sentence replaced by the ruled text verbatim, and the book rebuilt (561 pages, no dropped glyphs).
• a299ddf4: CHANGES.md row 75.
• SC-42: not regenerated; the output is byte-identical to the committed result.

SC-42

sc42-integration-result.json was regenerated with the unchanged checker.

• Checks: 678 → 689, which is 11 records added and 1 replaced, all passing.
• Only the checks array changed.
• Normal and -O runs are identical, and the self-test output equals the committed controls file.
• sha256 fc79539fb4d1dc4eef184320ee5f39b3f5a85ba02d273126abf4d6a78ed96565.

Checks

• Local: all OK.
  • coverage_check
  • voice_check
  • claims_check
  • duplicate_check
  • staleness_check
  • lean_manuscript_census
  • artifact_placement_check
  • audit_census_probe
  • SC-42 checker
• Build: 0 missing glyphs.
• CI: ordinary PR CI only; no workflow was dispatched.

Limits

• Not authorized: computation, Lean builds, dispatch, new premises, theorem promotion, merge, and frozen-record changes.
• ARROW-1: is the arrow of time derived, or inherited from time-asymmetric initial conditions? (cross-document disagreement) #1112 stays open.

🤖 Generated with Claude Code

https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

[amaybaum]

Independent scientific review of the ARROW-1 recovery note at exactly bd398f05af3d778c149db9d1a627f80efffd998e (base f5c281ae).

• This review is read-only and model-authored, by a separate agent. It is not an authenticated human review.
• No computations or Lean builds were run. Every derivation was checked by hand.
• Cai & Kim (arXiv:hep-th/0501055v1) and Wald (arXiv:gr-qc/0507094v1) were read first-hand.

Verdict: ACCEPT WITH CHANGES (two small blocking fixes)

The note's central conclusions hold:

• Alignment is open and unresolved, not refuted.
• The GSL is open.
• Theorem A.6 is a two-sided bound, not a monotonicity theorem.
• §4.6 selects where observers can exist, not a time direction.
• The horizon clock holds only as ambient FLRW kinematics under H>0 and the NEC, and is constant in exact de Sitter.

Items

1. §E.2: CONFIRMED.
   • The general-curvature form ṙ_A = 4πG·H·r_A³(ρ+p) and dA/dt = 32π²G·H·r_A⁴(ρ+p) check out, as does the flat form −8πc²Ḣ/H³. Dimensions check.
   • Two wording slips:
     • "non-decreasing exactly when H>0" fails when ρ+p≡0;
     • given H, the NEC fixes the sign, so "controls only whether nonzero" is wrong.
   • The ä/a equation is standard, not from Cai & Kim.
2. §E.3: CORRECTED (blocking).
   • The derivation is right: Ṙ_e = H R_e − c, and R_e ≥ c/H when Ḣ≤0 on [t,∞), with equality iff H is constant on [t,∞).
   • T1 is wrong for the event horizon. It imposes H>0 and the NEC only "on an interval I". The event horizon depends on the whole future: a later phantom phase approaching a big rip gives R_e→0, so Ṙ_e<0 somewhere on I.
   • Fix: either state the conditions on [t,∞), or restrict T1 to the apparent horizon.
3. §E.4: arithmetic CONFIRMED; scope CORRECTED (blocking).
   • The phantom signs are correct: H>0, Ḣ>0, ä>0, dA/dt<0.
   • But phantom violates the NEC, and GR excludes phantom crossing (conditionally; GR:12, :358, :553). Within the corpus's NEC background, ä>0 with H>0 implies dA/dt≥0. So "monotonic for as long as ä>0" is weakly true there: it is the wrong criterion and omits the decelerating era.
   • Genuinely refuted in-class: "S_dS increases" in exact de Sitter.
   • Fix: "refuted as a general kinematic criterion (phantom, NEC-violating), and false as strict increase in exact de Sitter; in the corpus's NEC background ä>0 is not the operative condition". This applies "state negative results only for the class actually ruled out".
4. §C.4: CONFIRMED.
   • φ(v,b) = (v⊕b, b) from δ₀⊗uniform gives the sector sums 0+ln2, then ln2+ln2, then ln2. Theorem A.6's hypotheses hold.
   • This refutes only the monotone reading. GR:607 (and GR:797's "non-decreasing") overstate.
5. §C.3: CONFIRMED, with a scope gap.
   • The identity M_k = H_{k+1}+H₂−H₁−H_{k+2} is correct and reversal-invariant under the counting measure.
   • Main:746 does not define P_Σ. Main's own examples fit the μ-orthogonal conditional expectation (self-adjoint), and then ‖(K†)ⁿ−Π‖ = ‖Kⁿ−Π‖. State this reading as an assumption.
   • Gap: "holds verbatim for φ⁻¹" also needs φ⁻¹ to have bounded range, since the proof uses φ's causal cone (Main:754, :364). A local bijection's inverse need not be local; x_i ↦ x_i+x_{i+1} mod 3 on an odd ring has a dense inverse. The gap closes under the T-invariance posit with a site-local T (SM Thm 17).
   • The conclusion "no direction" is unaffected.
6. §C.5: CONFIRMED.
   • The proof checks against SM Thm 19 (SM:641) and ch06:62.
   • State that A and S_cg are T-even.
   • Suggested witness: the swap φ(v,h) = (h,v) with uniform prior gives the visible process (a,b,a,b,…), which is reversal-symmetric with M₁ = 1 bit.
7. §C.2: CONFIRMED.
   • Main:518 is an existence theorem, so word it "the class realizable from a root is the same for backward runs".
   • Positive finding: the abstract step is kernel-checked as no_universal_oriented_property (OrientationClosure.lean:249). Cite it [K] for the abstract form.
8. §B.3: CONFIRMED. ρ_s = 3c²H²/(8πG) (GR:428) is instantaneous, even in H, and is a boundary energy density. It is not a relation between entropy and area.
9. §B.7: CONFIRMED.
   • An §A.19 search of the papers, the book and verification/lean* found no proof of Q1–Q4. The only monotone entropy-adjacent theorem is counting_passive, which takes Monotone Ω as a hypothesis.
   • There is no counterexample, so "open, not refuted" is correct.
   • The weak form (β) is accurate but trivial.
10. §B.10: CONFIRMED.
    • (EM), the iterated channel and the protocol orientation all appear in §5.4's own text, so the refutation of :413's "single open dependency" is legitimate.
    • Correction: ä>0 does not imply H>0. The dependency is the expanding branch implicit in "Λ-dominated/accelerating phase".
11. A11 (ch18:202): CONFIRMED, properly scoped.
12. §F.4, Main:758 retitle: CONFIRMED.
    • Main:744's "stated in the past-hypothesis corollary" would dangle; it is the only use of that label.
    • proof_census.slug would change MAIN:C-past-hypothesis-replacement-rate-condition to MAIN:C-observer-selection-rule-rate-condition, breaking coverage_check rules 1 and 2.
    • Both are necessary consequences under §A.25. Because the allowlist excludes them, they need a recorded owner ruling. Recommended scope:
      • the :744 label phrase only;
      • in LEDGER.json, id, label and fingerprint only, with kernel: GAP and the assumptions unchanged.
    • Main's independence from downstream papers is preserved.
    • Main:760(iii): the note names no defect there, so state one or leave it unchanged.
13. Locators: CONFIRMED (40+ spot-checked).
    • Write book/README.md:90.
    • Substratum:379's "that structure" is the FRW symmetry pattern, so it is weak support for B.2.

Classification check

• A1 is correctly labelled ambient. Its summary cell must say area, with the entropy reading open (GR:56).

• "Refuted" labels are sound for:

  • GR:607's monotone reading (counterexample);
  • A10 (text contradiction);
  • A11 (C.5 plus the witness).

  ä>0 needs the scoping above. None is applied to alignment, the GSL or the thermodynamic step.

• No corpus result proves A2–A7. A3 stays "open" for the relational target. Phase 2 must drop the literal universal "any horizon-bounded system increases".

• §A.31 in the framework's favour: GR's conditional exclusion of phantom crossing strengthens A1's conditional chain, and no_universal_oriented_property gives a kernel form for C.2.

Required revisions

Blocking:

1. T1, B.1 and §D: put the event-horizon hypotheses on [t,∞), or restrict T1 to the apparent horizon.
2. E.4, A1 and §D: scope "ä>0 refuted" as in item 3.

Non-blocking:
3. A1 summary: area only; entropy reading open.
4. C.2 and C.3: state the P_Σ reading and the bounded range of φ⁻¹.
5. E.2 wording.
6. B.10: ä>0 is not the expanding branch.
7. B.2: the Substratum:379 citation is weak.
8. E.1 citation list.
9. C.5: T-even assumption, and the optional swap witness.
10. C.2: Main:518 wording, and the optional [K] cite.
11. book/README.md:90.
12. Main:760(iii): state the defect or leave it unchanged.
13. ch18:208: the error is the dropped θ and the dropped hypotheses (partition-preserving T, counting prior), not "counts vs probabilities".
14. Phase-2 site gaps:
- ch02:192: "C2 … picks out one direction" belongs under A5. "This phase is the one in which S_dS is monotonically increasing" is alignment content.
- ch18:190: "within a single cycle, the observer's experience is unidirectional" conflicts with C.1. It is not on the allowlist, so it needs an owner ruling.
- ch04:32: "structural arrow of horizon growth" carries wording that treats the clock as an arrow.

Phase 2 readiness

After fixes 1–2, the note is a sound basis.

• F.1–F.3: supported.
  • Name the apparent horizon, with H>0 and the NEC as inputs.
  • Write "non-decreasing; strict while ρ+p/c²>0; constant in exact de Sitter".
  • Alignment is "not established", not refuted.
• F.4: pending the owner's ruling.

Reviewer's owner-relevant conclusion

• Proved:
  • exact recurrence, so no globally monotone state function;
  • §4.6 is a conditional, time-symmetric selection of where observers exist, given φ⁻¹ has bounded range; it supplies no direction;
  • Theorem A.6 is a two-sided bound;
  • under T-invariance and the counting measure, the traced process is reversal-symmetric up to θ.
• Conditional:
  • the monotone horizon area (ambient FLRW; H>0 and the NEC; the expanding branch is an input);
  • the thermodynamic, memory and measurement directions (non-stationary or product preparation, plus the orientation convention).
• Unresolved (open, not refuted):
  • alignment (Q1–Q4);
  • the GSL and the area–entropy identification;
  • exact-dynamics monotonicity;
  • why the expanding, low-entropy branch is realized.

Not checked:

• Physics Modulo Gauge;
• the source of the Ellis objection;
• Main reference [31];
• Carroll–Chen first-hand;
• FULL byte identity;
• a total-variation reading of τ_mix (the spectral/L² reading was taken);
• no Lean or computation was run.

---

Generated by Claude Code

[amaybaum]

Checkpoint 37b39d1c5bbea2efd4a10d9a8c69d127dd593bf5: Phase 1 revised to address the review in c.6649600806. This is model-authored by writer-arrow, and only RECOVERY.md and README.md changed. No manuscript was touched, and nothing was computed or built.

Blocking items

1. Event horizon (T1, B.1, §D). T1 is split:
   • T1a, apparent horizon: the conditions hold on the interval considered. This is the version Phase 2 relies on.
   • T1b, event horizon (flat): H > 0 and Ḣ ≤ 0 must hold on all of [t, ∞), with equality iff H is constant there. The reason is stated: a later phantom phase gives R_e → 0.
2. Scope of the ä > 0 verdict (E.4, A1, §D).
   • It is refuted as a general kinematic criterion: the phantom case, which violates the NEC.
   • It is false as strict increase in exact de Sitter.
   • In the corpus's NEC-respecting background, where GR excludes phantom crossing conditionally (GR:12, :358, :553), it is weakly true but not the operative condition, and it omits the decelerating era.
   • A "not necessary" row does not refute a sufficiency claim.

Non-blocking items

3. A1 summary: area only; the entropy reading is open (GR:56).
4. §C.3:
   • The P_Σ reading is stated as an assumption: the μ-orthogonal conditional expectation, self-adjoint, with ‖(K†)ⁿ−Π‖ = ‖Kⁿ−Π‖.
   • The bounded range of φ⁻¹ is stated as an assumption, with the counterexample x_i ↦ x_i + x_{i+1} mod 3.
   • That assumption holds under site-local T-invariance (SM Theorem 17, wave_time_reversal_invariant).
5. §E.2: the equality case ρ + p ≡ 0 is added, and the text now says that given H, the NEC fixes the sign.
6. §B.10: ä > 0 does not imply H > 0. The dependency is the expanding branch, which is implicit in the "Λ-dominated/accelerating phase" wording.
7. §B.2: the Substratum:379 citation is marked as weak support, since "that structure" there means the FRW symmetry pattern.
8. §E.1: the ä/a equation is now marked as standard, not taken from Cai & Kim.
9. §C.5: the assumption that A and S_cg are T-even is stated. The swap witness φ(v,h) = (h,v) is added: its visible process is reversal-symmetric with M₁ = 1 bit.
10. §C.2:
    • Main:518 is worded as "the class realizable from a root is the same for backward runs".
    • no_universal_oriented_property (OrientationClosure.lean:249) is cited as [K] for the abstract step, verified against the source. The φ ↦ φ⁻¹ instantiation is by hand.
11. Locator: book/README.md:90.
12. Main:760(iii): no change. It makes no monotonicity, direction or low-entropy claim, and Main:72 says Main's proofs do not use the cosmological realization.
13. ch18:208: the error is now stated as the dropped θ and the dropped hypotheses (a partition-preserving involutive T, the counting prior).
14. Phase 2 site list:
    • ch02:192 is reclassified: its C2 clause goes under memory/measurement, and its phase/growing-entropy clause under alignment.
    • ch04:32's "structural arrow of horizon growth" is moved from retain to conditional.

Pending owner rulings

These are outside the current allowlist and stay untouched until the owner rules:

1. Main:744's label phrase. It is the only use of the "past-hypothesis corollary" label, and it would dangle after the retitle.
2. verification/coverage/LEDGER.json. The id, label and fingerprint must change to MAIN:C-observer-selection-rule-rate-condition, keeping kernel: GAP and the assumptions unchanged.
3. ch18:190. "Within a single cycle, the observer's experience is unidirectional" conflicts with §C.1.

Noted, not changed

GR:797's "non-decreasing" is the definition of the generalized second law, followed by "supplies the following inequality toward it". I read it as correct and leave it to the reviewer. GR:607 remains the overstatement.

Phase 2 does not start until the owner rules on the items above.

---

Generated by Claude Code

[amaybaum]

Delta review of the ARROW-1 recovery note at exactly 37b39d1c5bbea2efd4a10d9a8c69d127dd593bf5. Compared against bd398f05; locators are at f5c281ae.

This is a read-only review written by a separate model agent, not a human review. Nothing was computed or built; every check was done by hand.

Verdict: ACCEPT WITH CHANGES (both blocking items fixed)

A. Blocking 1 (event horizon): ADDRESSED.

• T1a, apparent horizon (RECOVERY:143–146), is correct. dA/dt at time s depends only on H, a and ρ+p at s, so conditions on an interval I suffice.
• T1b, event horizon (RECOVERY:147–152, E.3:628–632), is correct. In flat space with H>0 and Ḣ≤0 on [t,∞), R_e ≥ c/H, so Ṙ_e ≥ 0. Equality holds iff H is constant on [t,∞).
• Phase 2 relies only on T1a (RECOVERY:153; F.2 at :702–707).
• Residual (non-blocking for Phase 2), in §D:570 and A1:35. These give T1b the apparent horizon's strictness and constancy conditions. For the event horizon, Ṙ_e(t) > 0 iff H is non-constant on [t,∞). Counterexample: an exact de Sitter stretch followed by an NEC-respecting decrease of H.
• Minor T1b wording at :148–150. "Somewhere" should be "on I". "Must hold" should be "are imposed on", because the condition is sufficient, not necessary.

B. Blocking 2 (ä>0 scope): ADDRESSED in E.4 (:663–676), A1, §D, the summary list and the review packet. "'Not necessary' does not refute a sufficiency claim" is correct.

C. Items 3–14: all ADDRESSED.

• The P_Σ reading and the bounded range of φ⁻¹ are stated as assumptions. The dense-inverse counterexample checks.
• no_universal_oriented_property is at OrientationClosure.lean:249. Its statement matches the quote; the abstract step is cited [K] and the instantiation [HA].
• The swap witness checks.
• Main:760(iii) is NO CHANGE, consistent with Main:72.
• ch18:208 is restated as the dropped θ and dropped hypotheses.
• ch02:192 is split three ways and ch04:32 is reclassified.
• Optional: wave_time_reversal_invariant (OI_Time_Reversal.lean:145) is stated for histories. The within-state layer swap is a hand translation and could be labelled that way.

D. The GR:797 disagreement: the writer is right. GR:797 defines the GSL ("The generalized second law is the statement that … is non-decreasing"). That is an open target, not an assertion. The earlier review's parenthetical about GR:797 is withdrawn. GR:607, "non-decreasing from a product initial state (Theorem A.6)", is the overstatement. The note treats both consistently.

E. Scope: CONFIRMED.

• Only the two record files changed.
• The owner's rulings (ARROW-1: is the arrow of time derived, or inherited from time-asymmetric initial conditions? (cross-document disagreement) #1112 c.6649643759, 04:25:02Z) postdate this commit (04:24:29Z), so items still marked "pending" is correct here.
• Phase 2's first commit should mark them authorized, citing c.6649643759.

F. New errors.

• Material: the T1b cells (see A).
• Minor:
  • RECOVERY:3–4 still says "Not yet independently reviewed";
  • README:154 points ch18:190 at §F.4, but its ground is §C.1;
  • the T1b wording noted in A.
• No new overclaim.

Required revisions

1. §D:570 and A1:35 (event horizon). State it as: non-decreasing; Ṙ_e(t) > 0 iff H is non-constant on [t,∞); constant iff exact de Sitter on [t,∞). Attach "strict iff ρ+p/c²>0" to the apparent horizon only. Not blocking for Phase 2.
2. Non-blocking.
   • Update the status at RECOVERY:3–4.
   • Fix the ch18:190 pointer.
   • Apply the T1b wording at :148–150.
   • Drop F.2's "leaves it to the reviewer".
   • Mark the three owner-ruled items authorized.

Phase 2 readiness

Yes. The note is a sound basis for Phase 2 under the owner's rulings.

• Phase 2 uses T1a with the apparent horizon named and H>0 and the NEC stated as inputs.
• The event horizon appears only with conditions on [t,∞).
• The ä>0 verdict is scoped to the classes it actually rules out.
• The scientific boundaries are respected:
  • alignment and the GSL remain open;
  • Theorem A.6 is two-sided;
  • §4.6 concerns observer eligibility, not direction;
  • the clock is conditional and constant in exact de Sitter;
  • the two horizons are kept distinct.

---

Generated by Claude Code

[amaybaum]

Checkpoint 24452b01231b868fa3d1ff077a8e8065920a7726: ARROW-1 Phase 2 manuscript edits pushed (WIP / UNVERIFIED). These are model-authored by writer-arrow under #1112 c.6649271094 and c.6649643759. They need the independent scientific and minimum-change reviews on this exact commit. This is not a request to merge.

Commits

• e133ffbc: record.
  • Applies the delta-review revisions (c.6649698455): the T1b event-horizon cells, the status line, the ch18:190 pointer to §C.1, the T1b wording, and GR:797 ruled as a definition.
  • Records the owner rulings.
• d3e13c99: Main and LEDGER.
  • Main:758 title becomes "Corollary (observer-selection rule, with its rate condition)"; its body and hypotheses are unchanged.
  • The Main:744 label phrase is updated.
  • In LEDGER.json, id, label and fingerprint become MAIN:C-observer-selection-rule-rate-condition / df04fba6b2d81e20. kernel: GAP and the assumptions are unchanged.
  • Main gains no downstream dependency.
• eb74b3e7: papers.
  • Substratum: abstract, :26, :56, §5.1, :379, and §5.4 at :401–415.
  • GR:32, :214, :561, :607, :709.
  • SM:1366, :1592.
  • Methodology:184, :227.
  • Explainer:930, :932.
• 4081cba0: book.
  • ch02:176, :184, :192, :194, :196, :198.
  • ch03:58, :202.
  • ch04:32.
  • ch18:190, :202, :208, :216.
  • Each edit is mirrored byte-for-byte into FULL.
• e4ab279a: regenerated .tex and .pdf.
• 24452b01: verification/audits/manuscript/arrow-1-20261105/CHANGES.md, a 51-row site-by-site table giving old and new text, class, RECOVERY section and mirror line for each edit.

Unlisted sites found and not edited

These are for an owner ruling; details are in CHANGES.md.

• ch18:214 (FULL:5027). "The thermodynamic arrow and the cosmological arrow … are aligned in the framework." This asserts alignment, and it sits next to the edited :216. The plan had grouped it as "already correct".
• Substratum:393 and :399. The "arrow-of-time projection" labels and the §5.4 heading.
• Substratum:409's heading, "What the framework derives".
• ch02:200 (FULL:843) and ch02:220 (FULL:863). Wording that frames the arrow as one of the three projections.
• GR:190. Already conditional; no change recommended.
• Methodology:212 and :214. These describe other authors' work.

Build

• Toolchain: pandoc 3.1.3, XeTeX TL2023.

• Commands: sh ./build.sh Substratum GR SM Methodology Explainer Main and sh ./build.sh --book.

• 0 Missing-character warnings.

• Page counts:

  Document	Pages
  Substratum	55
  GR	91
  SM	150
  Methodology	53
  Explainer	72 (was 71)
  Main	100
  Book	561 (was 560)

• Mean characters per page: 2786–3277.

Local checks

All exit 0, at the base and after the edits:

• coverage_check
• voice_check
• claims_check
• duplicate_check
• staleness_check
• lean_manuscript_census
• artifact_placement_check

CI

The ordinary PR verify run 41969128746 on this head was queued at checkpoint time. No workflow was dispatched.

---

Generated by Claude Code

[amaybaum]

Checkpoint bd926c76316173086a41ec5770ab80415155d598: CI follow-up for ARROW-1 Phase 2 (WIP / UNVERIFIED). Model-authored by writer-arrow.

Run 41969128746 at 24452b01 and run 41970039595 at 6010cf44 each had exactly one red job, "Numerical probes / core B". Every other job was green. The causes were checked locally on the branch.

1. audit_census_probe.py, AUDIT-B B1-required (run 41969128746).
   • The authorized ch18:216 correction adds one correct "not presently discharged" sentence about C4 in ch18 and in FULL. Counts: ch18 0 → 1, FULL 3 → 4.
   • At 6010cf44 I updated only expect and why in verification/audit-census.json, following that manifest's own drift rule. The pattern was not widened, and the probe passes.
   • The manifest is a mutable gate file, not a frozen record. It was not on the allowlist, so it is recorded in CHANGES.md for confirmation. It is easy to revert if the coordinator or owner prefers.
2. SC-42 integration step (run 41970039595).
   • (a) Fixed at bd926c76. The added-text caps scan hit "FLRW" in Substratum:405 (§A.33). It is now spelled out as "Friedmann-Robertson-Walker", as Substratum:379 already does, and Substratum is rebuilt at 55 pages with 0 missing glyphs. The checker now reports verdict: PASS with no failed predicates, and its self-test output equals the committed controls file.
   • (b) Not fixed; reported. The step also compares the checker's output byte for byte (cmp) against the committed sc42-integration-result.json of the bell-operational-reconciliation-20261029 round.
     • The checker adds one "chapter line mirrored in FULL" predicate for each chapter line changed since its base ca3a5bb2.
     • This round adds 10 such predicates and replaces 1. All of them pass.
     • Per the coordinator's instruction, I did not change the checker or its expected files. This step will stay red until the coordinator or owner decides. Precedent: d83f8723 (Correct the false Lemma-1 → total-S finiteness inference (minimal source correction) #1013) regenerated that file after a chapter edit.

Local checks at bd926c76 all pass: coverage_check, voice_check, claims_check, duplicate_check, staleness_check, lean_manuscript_census, artifact_placement_check, plus the census probe and the SC-42 verdict. No workflow was dispatched.

---

Generated by Claude Code

[amaybaum]

ARROW-1 Phase 2: independent scientific review at bd926c76316173086a41ec5770ab80415155d598 (base f5c281ae). This review is read-only and model-authored. Nothing was committed, pushed, dispatched or computed beyond hand algebra. Minimum change, mirrors, builds and CI are covered by the separate reviewer and are not repeated here.

Verdict: ACCEPT WITH CHANGES

The round meets its scientific aim. No edited site describes the arrow as derived, and alignment is stated as open, never as refuted. Main §4.6 is described as time-symmetric eligibility, the generalized second law stays open, and Theorem A.6 is stated as a two-sided bound. Five blocking items remain. Four are fixes of a few words on sites already in scope; the fifth (B4) needs an owner ruling.

Independently verified

• Substratum:405 kinematics. For flat FRW, $A=4\pi c^2/H^2$ gives $dA/dt=-8\pi c^2\dot H/H^3$. From $H^2=8\pi G\rho/3$ and $\dot\rho=-3H(\rho+p/c^2)$ it follows that $\dot H=-4\pi G(\rho+p/c^2)$, so $dA/dt=32\pi^2Gc^2(\rho+p/c^2)/H^3$. Given $H&gt;0$:
  • the area is non-decreasing under the NEC;
  • it is strictly increasing iff $\rho+p/c^2&gt;0$ (Λ contributes zero to this sum);
  • it is constant in exact de Sitter.
  • Matter and radiation give $\ddot a&lt;0$ with $dA/dt&gt;0$.
  • Phantom: $a\propto(-t)^n$ with $n=2/(3(1+w))&lt;0$ gives $H=n/t&gt;0$, $\dot H=-n/t^2&gt;0$ and $\ddot a/a=n(n-1)/t^2&gt;0$, so A shrinks while $\ddot a&gt;0$.
  • So "not $\ddot a&gt;0$" is correct, and so is "constant in exact de Sitter, where $\ddot a&gt;0$". ch02:192's version without "flat" also holds: by RECOVERY §E.2, $dA/dt\propto H\tilde r_A^4(\rho+p)$ for any curvature.
• Theorem A.6 is two-sided. GR:799 states the bound "for every t". The proof (subadditivity, then unitarity) uses no sign of t, and $U_\varphi$ is invertible. Substratum:407's "at every earlier and later time" and GR:607's restatement are therefore exactly supported. I checked the CNOT counterexample (RECOVERY §C.4): $\Sigma$ goes $\ln2\to2\ln2\to\ln2$. It refutes only the monotone reading.
• New citations resolve correctly.
  • [GR, §2.3] (GR:56): the calibration is on exact de Sitter, and the dynamical Kodama–Hayward extension "is not established here".
  • [GR, §7.2] (GR:428–430, also :388): $\rho_s=3c^2H^2/(8\pi G)=\rho_{\rm crit}$ is an instantaneous identity, even in H.
  • [GR, §A.6] (GR:793–805): Theorem A.6 and "the generalized second law is open".
• Substratum:411's replacement ("reduces to the ordering of moments together with φ") is supported by Substratum:413 and Methodology:184. The old Main §1.3 citation was wrong.
• GR:214 matches counting_passive (ThermalOrientation.lean:256), whose hypothesis is Monotone Ω, with energy passivity as the conclusion. Writing $\beta_E&gt;0$ implies that hypothesis.
• ch18:208 matches SM Theorem 19 (SM:641): an involutive T preserving the partition, the counting prior, $T_{ij}=T_{\theta j,\theta i}$ at every n, and $\theta=\mathrm{id}$ giving detailed balance.
• ch18:202's "under the invariant measure the trace-out alone produces none." I checked the RECOVERY §C.5 hand argument. $s'=\mathcal T\varphi^n s$ is μ-preserving, and $\varphi^k\mathcal T=\mathcal T\varphi^{-k}$ gives $\pi_V\varphi^k s'=\theta,\pi_V\varphi^{n-k}s$. The history law is therefore θ-reversal-symmetric. This holds under the T-symmetry the ch18 paragraph already states.
• ch18:216's C1–C3 "verified", C4 "not presently discharged" matches GR:46–52 and ch07:12, :30.
• ch18:190. By RECOVERY §C.1 (summing around a cycle), no state function can increase throughout a cycle. The new text asserts no reversal, as ruling item 2 requires.
• Main. The .md diff is exactly the :744 label phrase and the :758 title. The statement, hypotheses, body and :760(iii) are unchanged. No citation, premise or downstream reference is added, so Main's independence is intact. The LEDGER changes only id, label and fingerprint, and kernel: GAP is kept.
• Register (§A.32/§A.33). I scanned every added word in papers/*.md: no caps-emphasis token (FLRW is already gone), no history phrase, no reader instruction, no self-assessment.

Blocking

B1. The apparent horizon is credited to the "causal partition" and called frame-independent. Sites: Substratum:405, and ch02:194 with FULL:837. Both now name the apparent horizon and restrict to comoving observers, but both keep "The monotonicity is a feature of the causal partition rather than of any local frame".

• GR §2.3 (GR:56) says the causal definition is the event horizon, and the two horizons coincide only in exact de Sitter.
• $A=4\pi c^2/H^2$ is defined by the comoving FRW slicing, so "rather than of any local frame" contradicts "comoving".
• This conflicts with the owner boundary "Preserve the distinction between event-horizon and apparent-horizon conditions" (c.6649643759).
• Fix, either of:
  • replace the clause with one that claims only what the apparent horizon carries, e.g. "the monotonicity holds for every comoving observer of the homogeneous background";
  • or, if the event horizon is meant, name it and put the conditions on $[t,\infty)$ (RECOVERY §E.3).

B2. The null energy condition is dropped from the summary statements of the clock. On the expanding branch the area is non-decreasing only with the NEC; the phantom row of RECOVERY §E.4 shrinks A while H > 0. The condition is live in the corpus: GR excludes phantom crossing only conditionally (GR:12, :358, :553), and DESI DR2's apparent phantom crossing is a recorded tension. These sites state it without the NEC:

• Substratum:14 and :26 ("non-decreasing on the expanding cosmological branch");
• Substratum:371;
• Methodology:184 ("on the expanding cosmological branch, horizon growth supplies a non-decreasing parameter");
• ch03:58 / FULL:932 ("on the expanding branch, horizon growth supplies a non-decreasing clock");
• ch04:32 / FULL:1149 ("the expanding branch, taken as given, on which the horizon area does not decrease"). The NEC is a further commitment here.

Substratum:407 ("as above"), :409 ("given the inputs above"), :413 and ch02:192–196 carry the condition by context. Fix: add "with the null energy condition" (or equivalent) at each listed site.

B3. A surviving "derived arrow" in an edited paragraph. Methodology:184 still opens "…an ordering of moments, a causal topology, and a derived arrow of time…", two sentences before the round's own "is addressed". This breaks the owner boundary that the arrow is not described as derived. Fix: delete "derived".

B4. ch18:214 (FULL:5027) needs an owner ruling before landing. It still reads "The thermodynamic arrow and the cosmological arrow … are aligned in the framework" and "Time evolution corresponds to the horizon entropy increasing toward equilibrium". The edited :216 that follows says the tracking "is not established". The round therefore creates an adjacent assert-then-deny pair (§A.30). I agree with the writer's recommendation to give it the same treatment as :216. For example:

• drop "are aligned in the framework" or state that alignment is open;
• make the horizon-entropy clause conditional on the expanding branch and the NEC (constant in exact de Sitter).

B5. An ambient calculation is presented as projection output. RECOVERY B.1 classes T1 as [AMB]: "a statement about the background OI is applied to, not a derivation from (S, φ)". OI's covariant field equations are the open G3 step (GR §8.7).

• The manuscript does not carry that classification. Substratum:405 states $dA/dt$ without saying the Friedmann dynamics is the ambient background.
• Substratum:14 and :371 say the third projection, among "the framework's derivations [that] apply the same trace-out machinery to the same $(S,\varphi)$", supplies the parameter.
• This is the Code Review Rules category "ambient/background calculation → native OI prediction".
• Minimal fix: one clause at :405, e.g. "computed from the Friedmann equations of the background, taken as given ([GR, §8.7])". Optionally "on the expanding Friedmann-Robertson-Walker background" at :14 and :371.

Non-blocking

• N1, Substratum:405. "Both conditions are inputs:" then names the expanding branch and the entropy reading. The entropy reading is not a condition, and the NEC goes unaddressed. Suggest "Both are inputs …; the null energy condition is assumed of the matter content".
• N2, Substratum:405. Writing $\dot H=-4\pi G(\rho+p/c^2)$ would make the sign argument self-contained.
• N3, ch02:192 / FULL:835. It states $S_{\rm dS}(t)$ "assigned to it … is non-decreasing" without the de Sitter-calibration caveat that Substratum:405 carries (GR:56, dynamical extension not established).
• N4, ch18:202. "the observer's reduced description is not [T-symmetric]" is still unconditional, one sentence before the new qualifier. Suggest "is not, given such a condition".
• N5, Substratum:411. "carrier of the asymmetry" and "the cascade would halt at the thermodynamic layer" still presuppose a cascade that :407 now conditions on boundary data. This is mild.
• N6, Substratum:56. "is fixed by a low-entropy or product boundary condition" is slightly stronger than :407's "rests on". Monotone increase under the exact dynamics is itself unproved (RECOVERY B.3, P2). Suggest "rests on".
• N7, Substratum:405. "the horizon parameter orders the expanding branch": the ordering is only non-strict (it is constant in exact de Sitter). This is a nit.
• N8, SM:1592. "— with the horizon-clock-plus-observer-selection projection developed in [Substratum §5.4], whose alignment …" is grammatically loose but scientifically correct.

Unlisted sites (§A.25 step 6 re-grep)

Terms: arrow(s), aligned/align, derived/derives, second law, GSL, time direction, direction of time, past hypothesis, horizon/structural clock, cascade, unidirectional, accelerating phase, monotonically increasing. Scope: papers/*.md, book/*.md at bd926c76. Recommendations:

Site	Recommendation
ch18:214 / FULL:5027	Change (B4), same treatment as :216
Substratum:393, :399 headings	No change; the labels match the narrowed framing ("a third projection bears on…")
Substratum:409 heading "What the framework derives…"	Change "derives" → "establishes" (cheap, matches the body); no change is acceptable
ch02:200 / FULL:843 "co-derived"	No change; if B5 is fixed at the anchor, optionally note the ambient input
ch02:220 / FULL:863 "three large emergent structures"	Narrow to match ch02:184 (low priority)
GR:190	No change; consistent with the kernel and the GR:214 correction
Methodology:212, :214	No change; they describe Barbour's and Deleuze's derivations, not OI claims
Explainer:924–926 (new)	Owner ruling. :924 "the arrow … has to come from … the observer", then :926 "A structural theorem (Main §4.6) establishes exactly that". This attributes the arrow's origin to §4.6, against the owner boundary, and conflicts with the edited :932 ("fixed by boundary conditions, not derived"). Planned as "retain", but the retained content should be the exclusion, not "exactly that". Recommend narrowing "exactly that" to what §4.6 establishes.
ch18:206 "produces the arrow of time"	No change; it reads within :204's stated low-entropy start
ch18:20, :328, :334; ch00:101; README:90; ch07:153–157; GR:797, :805	No change; correct as written

Owner-ruling items

1. Add ch18:214 / FULL:5027 (B4).
2. Explainer:924–926 "establishes exactly that".
3. Optional: the Substratum:409 heading, and ch02:220.
4. Confirm the out-of-list consequences already recorded in CHANGES.md: audit-census.json B1-required and the SC-42 cmp. These are not scientific; they are deferred to the minimum-change/CI reviewer.

B1, B2, B3 and B5 fall inside sites already authorized, and each needs a few words.

---

Generated by Claude Code

[amaybaum]

Minimum-change, propagation, artifact and CI review of ARROW-1 at exact head bd926c76316173086a41ec5770ab80415155d598 (base f5c281ae, which is still main). This is an independent, read-only, model-authored review. I wrote nothing to the repository. The scientific content is out of scope here; a separate reviewer covers it.

Verdict: ACCEPT WITH CHANGES. The manuscript diff is exactly the change record, with no unmapped hunks. Mirrors, LEDGER and artifacts are correct. Two items must be settled before landing:

• the SC-42 expected-result file (item 5b below);
• an owner ruling on ch18:214 (finding F1).

1. Minimum change

• Every hunk maps to a row. I parsed all 51 rows of CHANGES.md and applied each old→new fragment to the base, with book rows also applied to FULL at the stated line.
  • Each old fragment occurs exactly once in its file, at the stated line.
  • The result is byte-identical to the head for all 11 edited .md files: Substratum, GR, SM, Methodology, Explainer, Main, ch02, ch03, ch04, ch18 and FULL.
  • No other .md file changed. There is no unmapped hunk.
• Every row maps to an authorized site. Each row's site is either in the plan c.6643840329 (plus the side corrections in c.6649271094) or added by ruling c.6649643759 (Main:744, ch18:190, ch02:192 and ch04:32).
• Retained sites are byte-untouched: Main:760, GR:797–805, Explainer:924–928, ch02:182, ch03:62, ch03:204, ch04:30, ch18:204/206/210/212/214/218–222/328/334. Complexity, Structure, ch00, ch07 and book/README.md are not in the diff.
• Main's diff is only the :744 label phrase ("past-hypothesis" → "observer-selection") and the :758 title. The corollary body and hypotheses are unchanged.

2. Mirrors (§A.14)

• At both base and head, every non-blank line of ch02, ch03, ch04 and ch18 equals the FULL line at the fixed chapter offset. So the 13 edited chapter paragraphs are byte-identical to their FULL paragraphs, and there is no drift elsewhere in those chapters.
• The FULL diff consists only of those 13 lines.

3. LEDGER.json

• Fields changed: only id (→ MAIN:C-observer-selection-rule-rate-condition), label and fingerprint. kernel: GAP, assumptions, area, line and checks are unchanged.
• Fingerprint: tools/proof_census.py --json at head yields the same id and label, and fingerprint df04fba6b2d81e20. That matches the ledger.
• tools/coverage_check.py: OK (134 canonical statements, 19 unattached checkers).
• Old id: survives only in frozen or historical audit records (SC34 README, post982 HUNKS), which is correct.
• Pre-existing, outside the ruling's field scope: the entry's line: 752 was already stale at base (the title is at :758). Not for this round.

4. Artifacts

• Exactly the right set. .tex/.pdf were regenerated for exactly the 6 edited papers and the book. Chapters have no artifacts.
• .tex matches source. Re-running pandoc locally on each .md reproduces the committed .tex byte for byte, apart from the trailing source-sha256 stamp. staleness_check OK (13 matched, 0 unstamped).
• Page counts (pdfinfo) match the writer's figures (base→head): Substratum 55→55, GR 91→91, SM 150→150, Methodology 53→53, Explainer 71→72, Main 100→100, Book 560→561.
• PDF text reflects the final edits:
  • Substratum.pdf contains "Friedmann-Robertson-Walker" and no "FLRW", so it was built after bd926c76.
  • Main.pdf carries the new title.
• Local checks at head, all exit 0: coverage, staleness, voice, claims, duplicate, lean_manuscript_census, artifact_placement_check and audit_census_probe.
  • legacy_records_check is UNDECIDABLE locally (shallow clone).
  • The Release gate step of the required Mathlib bridge job is green at bd926c76.

5. CI

• Run 41971221971 at bd926c76 (pull_request; in progress when read):
  • Green: Mathlib bridge (Release gate step success), Lean kernel check, and 12 probe jobs so far.
  • Red: "Numerical probes / core B", at step 8 "SC-42 integration source scope and must-fail controls".
  • Still running: the remaining probe jobs. The "Numerical probes" aggregate will be red because of core B.
• Run 41970039595 at 6010cf44: red at the same step. At that head it also had the FLRW caps hit, which bd926c76 fixed.

(a) verification/audit-census.json, B1-required. The change is correct.

• What the drift rule allows. The manifest's on_drift says: re-read the location, confirm the disposition, update expect with the reason in the commit message, and never widen the pattern.
• The diff follows it. It adds book/ch18-beyond.md: 1, moves FULL 3→4 and appends a provenance clause to why. Pattern and disposition are unchanged. audit_census_probe passes. The file is not in legacy-records.json.
• The new instance is correct usage. ch18:216 now reads "history readback C4 is a named realization condition not presently discharged (Chapter 7 §7.2)". That matches ch07:12/30 and GR:20.
• One procedural deviation (advisory). The rule says to update expect "in the same commit as the text change". The update came in 6010cf44, two commits after the text change in 4081cba0. The reason is in the commit message, and history must not be rewritten (§A.43.3), so no action is needed beyond noting it.

(b) SC-42 result file: regeneration is the correct remedy.

• Reproduced locally at head, in the step's own order:
  • normal --verbose exits 0;
  • normal output equals optimized output;
  • --self-test equals optimized, and both equal the committed sc42-integration-controls.json;
  • the only failure is cmp against the committed sc42-integration-result.json, "differ: byte 57775, line 1567". The job log shows the same byte and line (diagnostic only).
• The diff is +37/−1 lines and contains only new passing mirror predicates:
  • Checks go from 678 to 687, all ok: true. Verdict PASS, failed_predicates: [].
  • base, checker_sha256, site_table_sha256, the counts, mode and schema are unchanged.
  • 10 new "chapter line mirrored in FULL" records, each mapped to a line this round edits: ch02:176, 184, 194, 196, 198; ch03:202; ch04:32; ch18:202, 208, 216.
  • One of the 10 (ch02:194, "The horizon area provides a structural clock…") replaces the old ch02:194 record ("The cosmological horizon entropy provides…"). So the net change is +9 and 1 replaced.
  • The other three edited lines (ch02:192, ch03:58, ch18:190) were already changed since ca3a5bb2, and their 60-character predicate prefixes are unchanged.
• Why regeneration is right. The file is neither frozen nor a native-round record: it is not in legacy-records.json, and there is no receipt or seal. It has been regenerated for chapter edits repeatedly, including d83f8723 (Correct the false Lemma-1 → total-S finiteness inference (minimal source correction) #1013) and at least 8 earlier commits.
• What exactly would change:
  • Replace the file with the stdout of python3 -I -B verification/audits/manuscript/bell-operational-reconciliation-20261029/check_sc42_integration.py --verbose at head. The expected sha256 is bd65f7a032a0d58771d188f1af4172fcd8b4effedb209c324e2ab8718ab6f0d2; the current one is e1ce6846….
  • The controls file and the checker do not change.
  • It goes in a separate commit with the reason in its message, like the LEDGER and audit-census consequences.
• Authorization. The file is outside the authorized list, so the regeneration needs the coordinator's or owner's confirmation. Until then core B stays red.
• Wording nit in CHANGES.md. "add 10 such predicates and replace 1" should read "add 9 and replace 1" (678→687), or "10 new records, one replacing an existing record".

6. Register scan (§A.32/§A.33)

• I ran a word-level scan of the added text in papers/*.md (f5c281ae..bd926c76, 150 added chunks). It covered the register family, the history family (including "no longer", "formerly", "corrected", "revised", "now reads") and any [A-Z]{4,} token.
• 0 hits. The SC-42 checker's own added-text scan (base ca3a5bb2) also passes at head.
• Re-grep (§A.25 step 6) finds no stale target phrases in papers or book: "four aligned arrows", "resolves the arrow", "closes the Ellis", "derives the arrow", "all other arrows align", "none is concealed", "past-hypothesis replacement", "non-decreasing from a product", "You don't need to posit", "[Main, §1.3] already".
  • The surviving \ddot a > 0 at Substratum:405 is the corrected criterion sentence.

Findings

• F1 (needs an owner ruling before landing). ch18:214 (FULL:5027) still says the two arrows "are aligned in the framework" and "Time evolution corresponds to the horizon entropy increasing toward equilibrium". It sits directly before the corrected ch18:216, which says the tracking "is not established".
  • The two adjacent paragraphs now assert and deny the same thing (§A.30, §A.25).
  • The writer found this and correctly did not edit it, because it is outside the authorized sites. The plan had listed ch18:204–214 as already correct.
  • I support the writer's recommendation: give ch18:214 and its mirror the same treatment as :216.
• F2 (required for green CI). Regenerate sc42-integration-result.json as in 5(b), with the coordinator's or owner's confirmation.
• F3 (advisory).
  • The B1-required update was not in the same commit as the text change (5a).
  • The CHANGES.md predicate count wording (5b).
  • The pre-existing stale LEDGER line: 752.
• Writer's other unlisted sites (Substratum:393/399/409 heading, ch02:200/220, GR:190, Methodology:212/214): I agree they are framing, labels or attribution with no assertion contrary to the edits. They are left for the owner; ch02:220 is the only one worth a glance against ch02:184.

---

Generated by Claude Code

[amaybaum]

Checkpoint 98d048083e243ab328d6aec41b78367f03e9e1c7. The owner rulings in #1112 c.6650301410 are applied in f6f7dd13. a762d418 regenerates the SC-42 result: 678 → 689 checks (11 added, 1 replaced, all passing), sha256 fc79539f…. CHANGES.md is updated in 98d04808 (rows 52–74). Local checks are green. Ready for fresh reviews and CI; no merge.

---

Generated by Claude Code

[amaybaum]

ARROW-1 Phase 2: fresh scientific review at 98d048083e243ab328d6aec41b78367f03e9e1c7, focused on the delta from bd926c76 (f6f7dd13, a762d418, 98d04808). Rulings checked against #1112 c.6650301410. This review is read-only and model-authored, with no writes besides this comment. The SC-42 regeneration (a762d418) and CI belong to the minimum-change/CI reviewer and are not assessed here.

Verdict: ACCEPT WITH CHANGES (one blocking item, C1, at ch18:214)

Every earlier blocking item and every non-blocking item is applied. The delta adds no new scientific claim, and the register of the added paper text is clean (no caps-emphasis, history or reader-instruction phrases). Each changed chapter line is byte-identical in FULL: ch02:192/194/220 = FULL:835/837/863, ch03:58 = FULL:932, ch04:32 = FULL:1149, ch18:202/214 = FULL:5015/5027. All edits stay within their lines, so line numbers are unchanged.

1. Earlier findings

Item	Status	Site(s) at 98d04808
B1 apparent horizon credited to the causal partition	applied	Substratum:405 ("holds for every comoving observer of the homogeneous background"); ch02:194 / FULL:837 (the clause replaced by "standard background kinematics, not a prediction derived from the substratum")
B2 NEC in summaries	applied	Substratum:14, :26, :371; Methodology:184; ch03:58 / FULL:932; ch04:32 / FULL:1149
B3 "a derived arrow"	applied	Methodology:184 ("an arrow of time")
B4 ch18:214 alignment	applied (as A1)	ch18:214 / FULL:5027, first sentence; but see C1
B5 ambient → native	applied	Substratum:405 ("standard general-relativistic background kinematics … taken as given ([GR, §8.7]), not a prediction derived from the substratum"); :14, :26, :371 ("background kinematics"). [GR, §8.7] is apt: GR:701 says the nonlinear field equations are "not established" and makes "not a claim to have derived general relativity from the substratum".
N1	applied	Substratum:405 (the NEC is "assumed of the matter content")
N2	applied	Substratum:405: $\dot H=-4\pi G(\rho+p/c^2)\le0$. Rechecked: flat Friedmann plus continuity gives this exactly, and the NEC gives the sign.
N3	applied	ch02:192 / FULL:835 ("assigned to it at the de Sitter calibration")
N4	applied	ch18:202 / FULL:5015
N5	applied	Substratum:411 ("carrier of its orientation convention … could not state the memory-arrow step")
N6	applied	Substratum:56 ("rests on")
N7	applied	Substratum:405 ("orders the expanding branch non-strictly")
N8	applied	SM:1592

2. Rulings

• A1, ch18:214. "Whether the thermodynamic arrow and the cosmological arrow … are aligned is open in the framework: it is neither established nor refuted." This is correct and consistent with :216 ("not established"). The paragraph's second sentence still overstates; see C1.
• A2, Explainer:924–926.
  • :924 now reads "from the observer's description of the dynamics, given boundary conditions".
  • :926 reads "A structural theorem (Main §4.6) constrains the observer side of this — where observers can exist, not which way time runs".
  • Neither claims a direction or origin of the arrow, and both agree with :932 ("which way it points is fixed by boundary conditions, not derived"). Satisfied.
• A5, Substratum:409. The heading now reads "What the framework establishes". The body still separates three things: the §4.6 content, which is time-symmetric under its stated hypotheses; the convention the observer imposes; and the horizon parameter, established "given the inputs above". Satisfied.
• A5, ch02:220 / FULL:863. It now reads "establish its relationship to quantum mechanics and general relativity, and what it establishes about the arrow of time". This is minimal and matches ch02:176 and :184. Nit: "it" (the framework) has no close antecedent, but this is not substantive.

3. Horizon wording, site by site

Checked at Substratum:14, :26, :371, :379, :405, :407, :409, :413; Methodology:184; ch02:192, :194, :196; ch03:58; ch04:32.

• Horizon named. Each site names the apparent horizon, either itself or through its anchor (:407 "as above"; :409 and :413 "given the inputs above" and the NEC listed; ch02:196 after :192–:194).
• Background and conditions. Each names an expanding homogeneous GR or FRW background and the NEC.
• Non-decreasing, not strict. Each says "non-decreasing" or "does not decrease". Where strictness appears (Substratum:405, ch02:192), it carries its exact condition, ρ + p/c² > 0, together with "constant in exact de Sitter".
• Substratum:379 says "growth … with ρ + p/c² > 0; in exact de Sitter the horizon area is constant". That is strict growth under the strict condition, so it is correct.
• Substratum:405 states the background-kinematics, non-substratum status explicitly. ch02:194 and Methodology:184 mirror it.
• Residual nits (non-blocking):
  • Methodology:184's later inputs list "(the dynamics, the expanding branch, the boundary conditions …)" omits the NEC, which the same paragraph states earlier.
  • ch02:196's "The parameter is conditional on the expanding branch" leaves the NEC to context.

4. Ruling on the open point, ch18:214's second sentence: it overstates (C1, blocking)

"Time evolution corresponds to the horizon entropy increasing toward equilibrium, with the boundary modes accumulating correlations and the hidden sector approaching maximum disorder."

• (a) Conditions missing. It states horizon-entropy increase without the expanding background and the NEC, and without the de Sitter-calibration caveat on the entropy reading. That is the unconditional summary B2/N3 removed elsewhere.

• (b) It reasserts what was just called open. It ties the horizon entropy's growth to the hidden sector's disorder and the boundary correlations. That co-movement is the relation GR:805 says is not established ("a relation between horizon area and the hidden sectors' coarse-grained entropy"), and recovery-note B.3 (P3) and B.7 (Q2) treat it as open. It sits one sentence after "alignment is open … neither established nor refuted", so the paragraph asserts and then denies (§A.30).

• Minimal fix (one sentence, FULL:5027 identical):

  Given that initial condition, on an expanding background satisfying the null energy condition, the apparent-horizon area, and the horizon entropy assigned to it, does not decrease and approaches its de Sitter value; whether the boundary modes' correlations and the hidden sector's disorder grow with it is part of the open alignment question.

• Authority. This falls inside the existing rulings and needs no new one:

  • ch18:214 and its FULL mirror are an authorized site (A1);
  • B1/B2 directs "where the result is summarized, state its conditions … non-decreasing";
  • A1 requires the paragraph to read alignment as open, neither refuted nor established.

  If the owner reads A1 as covering only the first sentence, this item needs a one-line confirmation.

5. No new claims; no assert-then-deny elsewhere

The added text introduces no new physical premise, theorem or citation beyond [GR, §8.7], which resolves correctly. The changes from bd926c76 restate conditions, statuses and the standard Friedmann relation. I re-read every delta site for assert-then-deny. Apart from C1, none remains: Substratum:405's "orders … non-strictly; that the other arrows align with it is not established" asserts nothing it then withdraws, and neither do Explainer:924–932 or ch18:202/:216.

---

Generated by Claude Code

[amaybaum]

Fresh review of ARROW-1 at exact head 98d048083e243ab328d6aec41b78367f03e9e1c7: minimum change, propagation, generated artifacts and CI. It covers bd926c76..98d04808, i.e. f6f7dd13, a762d418 and 98d04808, against the owner rulings in #1112 c.6650301410. This is an independent, read-only, model-authored review: I wrote nothing to the repository, and a separate reviewer covers the scientific content.

Verdict: ACCEPT. CI at this head is fully green. The one deviation from the letter of ruling A3 is explained by rulings A1 and A5; see item 4.

1. Edits match the record

I applied CHANGES.md rows 52–74 as fragment replacements to the text at bd926c76, with each book row also applied to FULL at its stated line.

• Every old fragment is unique in its file and sits at the stated line.
• The result is byte-identical to the head for all 9 edited .md files: Substratum, SM, Methodology, Explainer, ch02, ch03, ch04, ch18 and FULL. No hunk is unmapped.

Each row maps to a ruling item:

Ruling item	Rows
A1	74
A2	66, 67
A5	60, 70
B1	58, 69
B2	52, 53, 54, 65, 71, 72
B3	64
B5	52–54, 56
N1	57
N2	56
N3	68
N4	73
N5	61, 62
N6	55
N7	59
N8	63

Every B2 site the scientific review listed is covered: Substratum:14, 26 and 371; Methodology:184; ch03:58; ch04:32.

Scope of the commits and record:

• f6f7dd13 touches only the 9 sources, their .tex/.pdf and CHANGES.md.
• 98d04808 touches only CHANGES.md.
• CHANGES.md says ch18:214's second sentence is left unchanged: A1 covers only the alignment sentence. That is within the ruling's scope.

2. Mirrors

• At head, every non-blank line of ch02, ch03, ch04 and ch18 equals the FULL line at the chapter's fixed offset.
• The FULL diff since bd926c76 is exactly 7 lines: 835, 837, 863, 932, 1149, 5015 and 5027. They correspond to ch02:192, 194, 220; ch03:58; ch04:32; ch18:202, 214.
• No drift anywhere else.

3. Artifacts

• Rebuilt set. .tex/.pdf changed for exactly Substratum, SM, Methodology, Explainer and the book. GR and Main .md/.tex/.pdf are untouched since bd926c76.
• .tex matches source. Re-running pandoc locally and appending the source-hash stamp reproduces each committed .tex byte for byte, stamp included. staleness_check OK (13 matched).
• Pages (pdfinfo): Substratum 55, SM 150, Methodology 53, Explainer 72, Book 561. These match CHANGES.md; GR 91 and Main 100 are unchanged.
• The PDFs carry the new text. For example:
  • "background kinematics" in Substratum (4) and Methodology (1);
  • "neither established nor refuted" in the book;
  • "constrains the observer side" in Explainer;
  • "where the alignment of the arrows" in SM;
  • no "a derived arrow" and no "establishes exactly that" anywhere.
• Local checks, all exit 0: coverage (134), staleness, voice, claims, duplicate, lean_manuscript_census, artifact_placement_check and audit_census_probe.
• No stale phrases left (re-grep). None of these survives in papers or book: "causal partition rather than", "derived arrow", "establishes exactly that", "are aligned in the framework", "cascade would halt", "carrier of the asymmetry", "horizon growth supplies".

4. A3: the SC-42 result

The regeneration is a separate commit. a762d418 changes only sc42-integration-result.json, +45/−1.

Untouched since base f5c281ae: the checker, sc42-integration-controls.json, sc42-intended-sites.json, validation.json, the record README, .github/ and tools/.

Reproduced at head:

• the normal --verbose run exits 0;
• normal equals -O;
• a fresh output equals the committed file;
• --self-test gives the same output in both modes, equal to the committed controls;
• sha256 is fc79539fb4d1dc4eef184320ee5f39b3f5a85ba02d273126abf4d6a78ed96565.

Only checks changed relative to the file at f5c281ae. base, checker_sha256, site_table_sha256, the counts, mode, schema and verdict are equal. The verdict is PASS, failed_predicates: [], and checks go from 678 to 689, all ok: true.

Every change is accounted for. There is one removed record, the old ch02:194 "The cosmological horizon entropy provides a structural clock…". There are 12 new records, all "chapter line mirrored in FULL", each matched to one round-edited chapter line:

• ch02: 176, 184, 194 (replacing the removed record), 196, 198, 220;
• ch03: 202;
• ch04: 32;
• ch18: 202, 208, 214, 216.

ch02:192, ch03:58 and ch18:190 are edited too, but add no new record: their 60-character prefixes were already recorded.

Deviation from the letter of A3. The ruling says "exactly nine added … and one replaced". The result is 11 added and 1 replaced.

• The two extra records come only from ch02:220 (A5) and ch18:214 (A1), which those same rulings authorized.
• The "nine" was the count at bd926c76, before those edits.
• The intent is met: only passing mirror records, no change to the tests. The owner should confirm the number when approving the merge.

5. A4: audit-census

• verification/audit-census.json and verification/coverage/LEDGER.json are byte-unchanged since bd926c76, so B1-required stays at ch18 1 and FULL 4.
• audit_census_probe: ALL CHECKS PASS.

6. CI: PR run 41973113384 at 98d04808, completed, conclusion success

• 25 jobs succeeded. They include "Mathlib bridge" (Release gate step success), "Lean kernel check", every probe shard, and the "Numerical probes" aggregate ("Require every probe shard" success).
• 1 skipped: "A42 exclusion (matrix)", which was skipped in every earlier run on this branch.
• "Numerical probes / core B": success. Step 8, "SC-42 integration source scope and must-fail controls", is success.
• This is a pull_request (synthetic-merge) run. main is still f5c281ae, the PR base.

7. Register (§A.32/§A.33)

• A word-level scan of the 33 text chunks added to papers/*.md since bd926c76 found 0 hits. It covered the register family, the history family ("no longer", "formerly", "corrected", "revised", "now reads" and similar) and any [A-Z]{4,} token.
• The SC-42 added-text scan, which runs from ca3a5bb2, also passes in CI.

---

Generated by Claude Code

[amaybaum]

Checkpoint (WIP / UNVERIFIED): ruling c.6651202921 applied at a299ddf49cd668ceafe22b9e769a4714717e9ed3 (fast-forward from 98d04808).

Commits:

• ca21185e:
  • In book/ch18-beyond.md:214 and book/The-Incompleteness-of-Observation-FULL.md:5027, the sentence C1 targets (c.6650406439 §4: "Time evolution corresponds to the horizon entropy increasing toward equilibrium, …") is replaced by the ruled text verbatim.
  • The two lines are byte-identical, and nothing else in either .md changes.
  • The book is rebuilt with sh ./build.sh --book: OK, no dropped glyphs, 561 pages (unchanged).
• a299ddf4: verification/audits/manuscript/arrow-1-20261105/CHANGES.md updated (row 75, grep, build and SC-42 results).

SC-42: no regeneration was needed. check_sc42_integration.py output is byte-identical to the committed result (PASS).

Local checks: legacy records (303 intact), citation roles, voice, claims, duplicate, staleness, coverage, artifact placement and census all exit 0.

Next: independent exact-head scientific and integration review, and CI at this head. No merge is authorized.

---

Generated by Claude Code

[amaybaum]

Exact-head review of a299ddf49cd668ceafe22b9e769a4714717e9ed3 (independent, read-only, model-authored; not the writer).

Scientific: ACCEPT.

• The right sentence was replaced. It is the one C1 quotes in §4 of c.6650406439. The ruling's "second sentence" takes C1's label; counted literally, the sentence is the third after the run-in heading. The ruling says it resolves C1, so the target is unambiguous.
• The new text is byte-exact to the ruling, and ch18:214 and FULL:5027 are identical.
• The new sentence is correct. For FRW of any curvature, R_A = c/√(H²+kc²/a²) and d(H²+kc²/a²)/dt = −8πG·H(ρ+p/c²) ≤ 0 when H > 0 under the NEC.
• The paragraph keeps every boundary:
  • alignment is open;
  • no unconditional horizon-entropy increase remains;
  • the entropy reading is conditional on calibration;
  • no approach to de Sitter is asserted.
• The other changed lines in the PR keep the same boundaries:
  • Main §4.6 constrains where observers can exist, not the direction of time;
  • GR Theorem A.6 is a two-sided bound;
  • the GSL is open;
  • apparent-horizon growth is labelled background kinematics.
• No overstated horizon-entropy or arrow claim remains anywhere in papers/ or book/.
• Optional, needs a new owner ruling to change the ruled text: "the stated calibration assumptions" has no antecedent in §18.8; a cross-reference such as "(Chapter 2 §2.6)" would make it resolvable. "FRW" is never spelled out in the book.
• Earlier non-blocking nits remain at Methodology:184, ch02:196 and ch02:220. None matters scientifically.

Integration: ACCEPT WITH CHANGES (PR metadata only; now applied to the description).

• .tex: pandoc reproduces the committed FULL .tex byte for byte, apart from the stamp. The stamp's source sha256 equals that of FULL.md.
• PDF: 561 pages; it contains the new text, and the old text has 0 hits.
• SC-42: unchanged and justified. A fresh run and the self-test are byte-identical to the committed files; the ch18:214 record is keyed by an unchanged 60-character prefix.
• Local checks: all pass (legacy records 303 intact, citation roles, voice, claims, duplicate, staleness, audit census).
• Mergeability: it was a fast-forward of f5c281ae.
• CHANGES.md: accurate. One optional note: lines 145–148 ("left as it stands") are superseded by row 75.

Pending:

• PR CI at this head.
• main has since moved to 53a3eb3d (#918 B1: retain the [CE]/[CR] evidence commits by ancestry only (history-only merge, no file changes) #1117, history-only, same tree). An ordinary merge of main and a delta check are needed before the merge gate.

No merge is authorized.

---

Generated by Claude Code

[amaybaum]

Merge record. Owner approval, amaybaum, Claude Code session chat, 2026-11-05:

#1113 ARROW-1 | 771f07a | APPROVED. […] Recheck ARROW-1 against the new main. If its reviewed content remains unchanged, it is conflict-free, and required checks remain green, merge it. Verify main CI again and record both merge results. For ARROW-1, I explicitly accept the SC-42 result of 11 added records and one replaced, rather than the earlier nine-record expectation. The two additional records correspond to authorized corrections. No new scientific claims or correctness-band promotions are authorized by these merges.

Recheck against the new main

• main had moved to 4f016afd (evidence(#1031 K1-RDY): written proofs of the blindness lemma, P1 and L-O(i)/(ii) (WIP / UNVERIFIED) #1119). An ordinary merge brought it into the branch, giving head c50089122559b6d073284d9dc169b8c2f60543ac.
• git diff 4f016afd c5008912 is byte-identical (cmp) to git diff 53a3eb3d 771f07ae. The reviewed content is unchanged: 31 files, +1961/−407.
• 303 legacy records are intact.
• PR CI run 41990050470 on c5008912: 25 checks passed and 1 was skipped. The Mathlib bridge Build and Release gate both succeeded. The PR was mergeable and clean.
• The PR was marked ready, then merged with "Create a merge commit" and expectedHeadSha.

Merge

• Merge commit: 37aafa5476fc7230b46a2c32b2a31c5c8c73fe70.
• Parents: 4f016afd (first) and c5008912 (second).
• Tree: 012ff2b0ad5f1f6333335fd3cf797b5c7642c231, identical to the head's tree.
• 303 legacy records are intact.

Post-merge CI. The verify push run 41991586727 on 37aafa54 succeeded: 25 jobs passed and 1 was skipped (the matrix placeholder). The Mathlib bridge Build and Release gate both succeeded.

SC-42. The owner accepts the record count of 11 added and 1 replaced (689 checks), as stated above.

Scope. This merge promotes no claim. The arrow-of-time question stays open under #1112. The ARROW-1 manuscript corrections, now on main, have to be accounted for before the #918 landing's final review, as D12 requires.

---

Generated by Claude Code