Skip to content

Audit cleanup inversion in the interleaved interface lemma - #33

Open
cursor[bot] wants to merge 1 commit into
mainfrom
research/interleaved-interface-h0
Open

cursor[bot] wants to merge 1 commit into
mainfrom
research/interleaved-interface-h0

Conversation

@cursor

@cursor cursor Bot commented Nov 15, 2026

Copy link
Copy Markdown

Change and mathematical scope

Audit, not a construction. The five named hypotheses H1–H5 of the interleaved scalar-interface lemma do not imply that cleanup inverts the raw z-action. Adding H0z — every z-destination gate in A and B omits its own destination — restores that implication for every finite word in the XOR-gate model. H0z and H1–H5 still do not imply the shear y + x.

The minimal witness is one source, one target and one helper: A is h ^= h, B is y ^= x, and C repeats h ^= h. H1–H5 hold and the helper mask stays 0.

This is not an acceptance bug in xcheck_p.py. replay rejects that gate before it reports H1–H5. A word that reaches the checker's PASS line already satisfies a stronger distinct-operand guard, which implies H0z. No scalar saving, paid-profile change, or κ improvement. The selected frontier is unchanged.

Remaining obligations: frame climbs in the same pinned proof paragraph, signed integer shears, the all-size compiler, and any formalization that transcribes only H1–H5. The selected 111-port word was not replayed.

Task and overlap

Base ef5f39403610d231a0c27e8a6b4f08c10209bcd6. Selected-record sha256 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32. Assignment seed 33969db763dc539fd8b92c2bb4cf1cc6, track adversarial-audit, card uniform-transfer-audit, scope key 603f341b99f590f17a76, priority revision 2.

Scope is refined. The card allows one restoration lemma. The object is the cleanup-inverse step in the pinned PR CrocSwap#543 text at research/e8-bit-r420/README.md (blob 6751eba6f617292fc919e58d5a9f9f06ab037391), not another finite supplier replay.

Refreshed overlap on 2026-11-15T12:20:51Z covered 913 open PRs and 7519 discussions, with complete pagination. Searching the parenthetical and "cleanup inversion" matched that lemma text and this package only. The nearest different audit is draft PR CrocSwap#2721 (head 6b14e212a1aba401d29f2bf7378e745ee2aed7f2), which uses the same z-gate induction for H2-sem condition 2. It does not test this operand parenthetical, and this audit does not contradict it.

Validation

python3 -B research/interleaved-interface-h0/verify.py
python3 scripts/research_package.py --package research/interleaved-interface-h0 --check-only

Observed: PASS_INTERLEAVED_INTERFACE_H0 and PASS_PACKAGE_METADATA_AND_PINS. Python 3.12.3, PYTHONOPTIMIZE unset, about 48 seconds. The same verifier under python3 -O also passes because the checks raise exceptions. The proof is the quantified argument in the README; the enumerations (1,507,142 / 343 / 1,755,904 / 21,296 words) are corroboration, and 208 short words satisfy H1–H5 while cleanup fails.

Not run: make frontier1094-verify, xcheck_p.py on a pinned word, native C++ replays, and frame or signed-integer proofs. g++ 13.3.0 is the c++ alternative and libboost-dev 1.83 is installed. The tuned requirements resolved to numpy 2.5.4, sympy 1.14.0 and scipy 1.18.1 and are not imported here.

Credit

The lemma and the involution parenthetical are eumemic's, from PR CrocSwap#543. The operand guard is the punctured checker's, in research/e8-bit-p111-tuned/code/xcheck_p.py. PR CrocSwap#425's interface lemma is Zhihao Chen's. PR CrocSwap#2721 is geckods's separate audit of this lemma. Apache-2.0.

Substantial AI assistance was used to draft the proof, the independent checker, and the package notes. This is not independent human review.

One Discussion createDiscussion attempt returned FORBIDDEN (Resource not accessible by integration) and was not retried.

Open next questions

The useful follow-up is the frame paragraph of the same pinned proof: whether one full frame and nested climbs make C legal for every H0z word, not only the selected word. This package does not load those frames.

A second question is whether a formalization transcribes only H1–H5. No Lean build was run.

Workflow feedback (optional)

Task uniform-transfer-audit, priority revision 2. Outcome: the five named hypotheses are not sufficient for cleanup inversion, and H0z repairs that step. The largest friction was the broad overlap vocabulary ("precision", "hosting"); searching the lemma's own parenthetical separated this question from PR CrocSwap#2721. One improvement would be a catalogue overlap term for the named interface hypotheses, so a restoration audit is not matched only through those generic words.

Open in Web View Automation 

H1-H5 do not imply that cleanup inverts the raw z-action. H0z, that no z-gate lists its own destination, repairs the implication in the XOR-gate model. The selected frontier and kappa are unchanged.

Co-authored-by: riquelme.pereira <riquelme.pereira@amentoriaenem.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant