Tooling: compile reader tables for a 1.40x faster PR845 replay

[theon199]

Change and mathematical scope

Tooling improvement; no new κ. PR845's complete P112 single-lift verifier repeatedly evaluates the same reader forms. This adds an opt-in engine that compiles the exact eight-by-112 table with the original evaluator, then generates every original ordered constraint row. The frozen package is unchanged.

Five alternating fresh-process pairs on macOS arm64 / CPython 3.12.13 measure median complete worker time 1.965451 s → 1.402593 s (1.401×). Untimed full replays count 979,776 → 896 targeted exact evaluations. The entire original certificate agrees: 511 support certificates, 3,272 sign cases, 128 conflicting kernels and nine inherited controls. Ordered row hashes, minors, kernels, conflicts and control outputs are compared; rank/histogram agreement alone does not pass.

Only research/p112-reader-memo-seed92/ is added. The cache snapshots complete ordered bases and labels, never caches X or coefficients, and holds at most four tables. Other input envelopes retain the original builder. No persisted reuse or skipped checks.

Task and overlap

• Exact base: 9ef085e6c07c8590329ad00b7eac382aef06124c.
• Contribution: 9f5e86a3458e57f5df62d2930232711d197790bf.
• PR845 reviewed source head: 125ce25bdae421627b56077adc14940c2d0aa52f; eight live Git-blob/SHA256 pins and historical provenance are in SOURCES.json.
• OS assignment seed 92, verification engineering; replay-runtime card, scope refined to this one kernel. Task and overlap record, sprint claim.

The final paginated metadata check retrieved 531 open PRs and 108 Discussion threads with no truncated comment/reply pages. A subsequent refresh screened the 50 most recently updated open PRs and eight latest sprint comments. No matching kernel optimization was identified; this is bounded overlap evidence, not a novelty guarantee. #1144 accelerates a different signed determinant checker, #1584 a finite-bit edge scorer, and #1452/#1595 source reads. #1626's scope map and the new #1656 decoder experiment are also distinct; no pending code is imported.

Refreshed main 3d161fdcdf9b49812efe3546e0ac5bff63f4de46 leaves all live inputs and the selected record unchanged.

Validation

python3.12 -B scripts/research_package.py --package research/p112-reader-memo-seed92

Expected: PASS_REGISTERED_RESEARCH_PACKAGE research/p112-reader-memo-seed92 outcome=completed.

This passes locally, including both full engines, 13 additional controls, five focused test groups and PR845's unchanged independent integer-minor/set-elimination checker. Controls detect mutated labels/bases, X/coefficient changes, cache eviction, integer-equal float failures, erased reader values, omitted/reordered constraints and source pins, and type-altered receipts. Tests cover proof-section corruption, lazy fallback, source drift, duplicate pins/JSON, actual optimized workers, read-only relocation with spaces and preservation of output after failure.

The five test groups also pass under optimized CPython 3.14.5. All 20 repository research-workflow tests and selected-record integrity pass. The committed tree passes a fresh exported replay under -I -O -B; staged whitespace checks pass. README gives single-engine and benchmark commands; BENCHMARK.json includes raw pairs, environment and excluded pilot runs.

Finite replay equivalence is not a formal theorem proof. Full selected cold replay, native/Lean builds and hosted CI were not run. The selected frontier and all-size/compiler, realization, hosting, precision/tape, analytic and bit-interface obligations are unchanged. Timing covers this worker only, excluding parent comparison/controls; no multiplication-runtime claim.

Credit

Credit Mihir Srivastava / @mihir-s-05 (PR845) for the theorem, geometry, certificate and both original checkers; DreamingOfClouds (PR717/446) for geometry and the inherited physical interpretation; Joel Pulikkan / GamingPuzzled (PR620) for coordinate cost 64 and its retained lineage. The independent-method checker remains predecessor work, not a new implementation claimed here.

Prepared for Sricharan Vagvala / Charan (@theon199) with substantial OpenAI Codex assistance in profiling, implementation, verification and writing. Parent-only execution; no independent human review. Apache-2.0; original sources, notices and upstream/ remain unchanged.

Open next questions

Review whether to adopt the compiled table in PR845's frozen package. Start with the registered replay and paired benchmark, then explicitly migrate source pins if approved. Maintainer review and that migration block adoption; this contribution remains opt-in. No merge, closure or frontier promotion is requested.

[Mister-Meeseeks]

Batch review at 9f5e86a3458e57f5df62d2930232711d197790bf.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Source-pinned PR845 reader-table memoization snapshots ordered bases/labels, excludes type aliases and retains uncached coefficients/X. Full certificate differential replay, failure/isolation controls and unchanged independent checker; opt-in tooling, no theorem change.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.