[aliromman]

This verification optimization makes the complete E7 minimum-support classification replay 17.53 times faster locally, with every scientific result and downstream check retained. Median time changes from 13.816 seconds to 0.788 seconds in six alternating fresh Python 3.12.14 processes. All outputs are byte-identical. Complete kappa is unchanged.

Correctness proof, exact bit-lane engine, scientific receipt, timing receipt, source pins, and validation and limits.

The predecessor loops over 134,217,727 nonzero words. The new method places coefficient assignments in integer bit lanes, adds their coordinate bits with exact Boolean adders, and selects all weights at most twelve. Every assignment is still covered. It returns the same 1,638 minimum supports, the same 630/1,008 reflection orbits, and the same exact rational classification. The iterator is proved for arbitrary independent binary generators, with bounded batch storage and no result cache. It remains exponential in code dimension.

PR #5669, exact head 700efb538a3d919409c9506d117846b7895b02ea, supplies the complete consumer. The derived verifier changes only its source-root location and minimum-support function. A syntax comparison checks every other statement. All reflection, frame, rank, ambient-pivot, omission and rejection checks run, and every source-receipt field must match. Frozen sources are not modified or patched at runtime.

python3.12 -B scripts/research_package.py --package research/e7-bitsliced-exhaustion

Expected markers:

PASS_E7_BITSLICED_EXHAUSTION
PASS_REGISTERED_RESEARCH_PACKAGE

Python 3.12 and the standard library suffice. The independent oracle checks all 1,381 unordered independent generator sets in F2^4, all five weight bounds and five batch sizes: 34,525 comparisons. There are also 108 wide-code comparisons through width 127, sixteen malformed-input and invalidation controls, and a second full E7 basis. Normal registered replay, optimized read-only Git-free replay with empty PATH, actual source-corruption rejection, committed-byte checks and official export integrity pass. All 24 export files retain their bytes and permissions. The performance measurement compares the complete scientific replay; the additional generic-code conformance suite is outside that timing comparison.

This reduces verification cost for E7/E8 deletion and decoder research. It supplies no new support theorem, physical supplier, paid-profile change or multiplication speedup. Full selected cold replay, native/outer admission, independent human review and proof-kernel acceptance are not claimed. Full-history publication preflight is unavailable in the shallow checkout; exact retained bytes and immediate ancestry are checked. Hosted CI is separate; existing PR #5659 and PR #5703 cover the shared main navigation failure.

Credit @petalas for PR #5238's geometry, frame code and support theorem, and @aliromman for the complete classification in PR #5669 linked above. The retained source notice preserves JoyboyBrian and the full E8 predecessor lineage. Boolean adders and bit-parallel enumeration are established methods; no literature-priority claim is made. New implementation, proof and independent small-code oracle: Ali Romman with substantial OpenAI Codex assistance, no subagents. Apache-2.0. Same assistant family as the consumed classifier; this is not independent human review.

Base and final main: df0137efd35e11b4445734bf612bdd4c7ebc330c, priority 7, selected conditional kappa .00310045. Source-retention commit: 1b853397390a537a3d63ed22976f8cc699743619. Final head: 337a65491f13076ba60225ef430554d0eacc2240. Canonical claim. Final overlap covers all 7,676 main notes, 404 open PRs and 5,072 Discussion records; no competing E7 bit-lane optimization was found within that coverage.