[amaybaum]

OI-BH-1: a bounded black-hole information audit based on book/ch07-gravity.md §7.6 and papers/GR.md Appendix A. It is exploratory research. It changes no manuscript and no existing proof module, and it makes no claim about physical black holes.

• Head: 84aca6b06cc23976e4e95c896fcc0fff988f01e4.
• Merge: not authorized.

Packages

There are two finite audit packages under verification/programmes/black-hole-information/.

bh01_access_v1: an exact classical access model. The tested finite classes are all bijections of small configuration spaces and named families, not the substratum dynamics. In them:

• bijectivity neither implies nor is necessary for recovery from accessible records;
• with bijectivity, complete endpoint access suffices;
• an inaccessible endpoint set K bounds the loss by P_succ ≥ 1/|K|.

Item F is an audit observation, not a model claim. By GR's own formulas, the Appendix A.7 premise τ_evap ≪ τ_B^D fails at 30 M_sun; that premise is still in main at GR:811.

bh02_scramble_v1.

• Permutation unitaries on configuration bases create no entanglement from configuration states, across bipartitions aligned with configuration coordinates.
• Page-like values arise only from superpositions the analyst supplies.
• t_P/t_evap = 1 − 2^{−3/2} holds conditional on β = 1, a constant mass-loss coefficient and S ∝ M². With Page's β ≈ 1.4847 it is 0.538 (Page 2013, Eqs. (2), (8)).

What it does not show

There is no statement about physical black holes, Hawking radiation, scrambling times, islands or the generalized second law. The physical inputs used are attributed to their sources: Page 2013's β in bh02 C7, and GR Appendix A.7's evaporation law in bh01 item F.

BH-03 and BH-05 to BH-07 are planned in PROGRAMME.md and have not been run.

The MANUSCRIPT-AUDIT.md files are proposals against the text at 809fac52. #920 has since corrected that text on main (merge c37700ad), independently of this PR, so the proposals must be re-checked against main before any use.

Review and landing audit

Each package keeps its separate-agent internal reviews in review/:

• bh01_access_v1/review/:
  • the review of 587d4196;
  • the written re-checks of 95458016 and df2ea6f5, with their summaries (redactions recorded).
• bh02_scramble_v1/review/REVIEW.md: the review of 1739f4e0 and the re-checks of 91e2230a, 1f1aec6f and ce7937d2.

A separate-agent internal landing audit of f6698781 was READY WITH NITS, and its re-checks of 606c98d8 and 84aca6b0 accept, with the description fixes applied here. It confirmed:

• claim separation;
• non-vacuous controls (bh01 mutants re-run, 12/12 caught; bh02 targeted mutants reproduced, full campaign recorded at 91e2230a);
• byte-identical replays and every SHA256SUMS entry;
• a clean merge with main;
• the cheap gate steps:
  • legacy-records: 303 intact;
  • V3 self-test and corpus;
  • 52 receipts hold.

Its nits are fixed in 606c98d8:

• reviews retained;
• manuscript-audit headers;
• Page time stated at β = 1;
• programme status;
• the C7 attribution to Page 2013;
• replay recipes.

The re-check nits are fixed in 84aca6b0: the written re-checks retained, the redaction record completed, and a provenance entry added. Across 606c98d8 and 84aca6b0, only the /C7/verdict leaf of bh02 result.json changes, and bh01's outputs are unchanged. Replays are byte-identical under -I -B and -I -B -O, and the targeted mutants reproduce.

Landing

Land with an ordinary merge commit, as #920 was. The packages cite plan, review and preregistration commits that exist only on this branch, and a squash or rebase would leave them unreachable from main.

Related: #920 (merged; corrected ch07 §7.6 and GR Appendix A without using this PR), #918.

🤖 Generated with Claude Code

https://claude.ai/code/session_013T6pfKuv5mi3MXtKjH2tZb

[amaybaum]

Writer handoff for #917.

At the owner's direction, the Claude session that writes #910–#912 is now the sole writer of this branch (research/oi-bh-information-audit). The existing draft is preserved as is: head 809fac52cdf207d24eccc63c734996bcb9d073a0, one commit adding verification/programmes/black-hole-information/PROGRAMME.md. Any other writer should stop modifying this branch; if a competing push appears, this session stops rather than racing it.

Planned work, local only until reviewed and separately authorized for push:

• BH-01: information recovery from emitted records under a fixed observer-access class, with one positive model and one reversible counterexample.
• BH-02: exact finite comparison of permutation-only and quantum scrambling models, with the Page-curve control and the supplied-ensemble distinction stated explicitly.

Each round commits its plan before computation and gets an independent review. No push, manuscript change, new physical postulate or merge is authorized by this notice.

---

Generated by Claude Code