Skip to content

Audit: in-repo HPE checkers do not read the pre-wrapper recipe - #2706

Merged
Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
hevlyo:research/hpe-prewrapper-binding
Oct 16, 2026
Merged

Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
hevlyo:research/hpe-prewrapper-binding

Conversation

@hevlyo

@hevlyo hevlyo commented Oct 15, 2026

Copy link
Copy Markdown
Contributor

Summary

  • States the boundary lemma already proved as factor_iff_pre: after (N), the wrapped block equals C₇ ⊗ id if and only if the executed pre-wrapper recipe R' equals the certified recipe R.
  • Shows that scripts/verify_hpe_lean_binding.py and research/hyperplane-end-lean/code/check_hpe_pairs.py never read that recipe. A synthetic 38-helper word whose records name E:FULL is accepted by both, and deleting the record is also accepted.
  • A changed adapter phase and an h += p gate are rejected, so the checkers do see monomials and the tail coefficient. They do not discharge hrec in hpe_word_exact.

No new kappa and no selected-frontier change. Seed d9b349359a3c923c0212b9514fe130b3, scope key e3ffea67887eeb4c396c, base b5ea56e0a5ac4552166253b1fa4be0753bb38c30. Claim: #2703

Overlap refresh on 2026-10-15 covered 782 open PRs and 1329 discussions. Neighboring work is the finished semantic port of the three Clifford pairs (#1394, PR #1415) and the HPE axiom-probe audit (#1593). Neither is this checker-versus-hrec gap.

Test plan

  • python3 -B research/hpe-prewrapper-binding/verify.py prints PASS_HPE_PREWRAPPER_BINDING full_recipe_accepted=true record_optional=true phase_rejected=true coefficient_rejected=true pairs=3
  • python3 -B scripts/research_package.py --package research/hpe-prewrapper-binding --check-only
  • Python 3.12 registered replay: PASS_REGISTERED_RESEARCH_PACKAGE research/hpe-prewrapper-binding outcome=negative-result
  • Selected baseline was not rerun: this sparse checkout lacks research/community-frontier1094-audit/ASSEMBLY.json. Selected record sha256 remains 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32.

Limits

The witness is synthetic, not an emitted word. The external control hpe_pre_wrapper_resolved_to_full_recipe is not rerun and is not claimed to be absent from the HPE tools. Footprints, well-labelled prefixes and the unit row stay assumptions. Predecessor credit stays with Jeremy Kothe.

Prepared by Riquelme Soares (@hevlyo) with substantial AI assistance. Apache-2.0.

Made with Cursor

factor_iff_pre needs the executed recipe to equal the certified (E, U) monomial before hpe_word_exact can drop the factorization hypothesis. Both shipped checkers accept a synthetic word whose pre-wrapper records name the FULL key.
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.

2 participants