Replays of the verifiers of #207, #202, #197 and #199 (no new κ)

[antoine-olivier]

I re-ran the verifiers of four open PRs from fresh clones at their current heads. #207 reproduces exactly, and so do #197 and the #199 diagnostic. #202's bit package passes, but its headline verifier doesn't reproduce here, because of a single environment-sensitive hash. No new κ, and nothing changes outside the new folder research/independent-reproduction-20261009/.

PR	Head	Verifier	Exit	Printed κ	Result
#207	cd14825023b7	coordinated-crossover-pr200/verify.py	0	1366380910073/2·10^15	reproduced
#202	8d8d67bcf69c	paired-cube-diagonal-bit-168/verify.py	0	166261725903847/2.5·10^17 (package's own κ)	reproduced
#202	8d8d67bcf69c	source-assisted-v4/verify.py	non-zero	none	not reproduced
#197	8c5e1cf07c23	packed-source-assisted-bit/verify.py … --full	0	135216063303877/2·10^17	reproduced
#199	5c1fb8346615	bit_reuse_maximal.py	0	diagnostic, output = committed	reproduced

All runs: Linux x86_64 (WSL2, Ubuntu 24.04.5), CPython 3.13.16, numpy 2.3.5, scipy 1.17.0, fresh clones of pull/<n>/head, unmodified sources, python -B. One control run used CPython 3.12.3. Heads re-read at 18:05Z today, unchanged.

For #202, source-assisted-v4/verify.py stops with Canonical certificate does not reproduce. The 7 leaves that differ all derive from one value: the hash of the gzipped complex lift certificate, written by exact_complex_flow_lift.py:409 (gzip.compress(..., compresslevel=9, mtime=0)). κ = 6768823/10^10 and every other leaf match. The lift I build is byte-identical to the one #194 committed, and #202 is #194 plus one commit (8d8d67b). #194's CI passed this same check under CPython 3.13.16, numpy 2.3.5 and scipy 1.17.0 (run 37931952877), which is my setup. #202's CI hasn't run: the PR has a merge conflict, and GitHub doesn't run pull_request workflows in that state, although source-assisted-v4.yml would run the same check. #194's check also fails here under CPython 3.12.3, so this check depends on the environment (interpreter or zlib build). I haven't isolated the cause and this isn't a defect claim. Recording the reference environment, or comparing the lift through its content as #207's verify_inner.py:23,29-30 already does, would avoid it.

For #197, the full regeneration exits 0, prints the claimed κ and leaves the tree clean. The removal of the 6,600 rank-60 exterior children rests on the paper proof of PROOF.md §4 (lines 81-93), an identity on arbitrary bank contents; the finite controls of lines 95-98 support it but don't simulate the full cover. The maintainer review of the comparable removal for #186 is research/community-round8-audit/BANK-SCHEDULE.md on main. I didn't check that #197's banks match that schedule. #205 removes the same children and #207 cites #197 for its bootstrap; I didn't assess whether their own bank checks cover this.

Not done: #205, #206, #208 and #199's new composition files weren't replayed, and no make verify ran, since the PR adds only Markdown, JSON and a standalone script. These are finite replays, not a formal verification: the conditional interfaces stay as each PR states them.

I did this with help from Claude (Anthropic) agents; the runs were done locally.

🤖 Generated with Claude Code

[Mister-Meeseeks]

Thank you @antoine-olivier. This contribution is now retained on main in the scoped 100-PR review and contributor credits.

Contribution: Independent Linux reproductions and portability diagnoses.

Review and limits: All changed files already match baseline main. Retain the author's environment and reports as independent reproduction evidence, not as executions performed in this round.

The complete original source, notices and relevant execution evidence are retained through the pinned source archive. The review distinguishes fresh local runs, source-bound historical Linux CI and prior exact-head review; it does not claim 100 fresh full mathematical audits. Closing this historical PR as reconciled. This preserves parallel and cumulative credit without replacing the selected frontier or promoting priced research targets to completed constructions.

Reviewed head: 0d584fad684b7999ae94edc6a1914901bdc9efa7. Publication: 50d768818f579d4e165b4c257c6ee25bca300802.