Count byte-language entrants missing from the generality board
perfloop-oss/casei · UNCATALOGUED MECHANISM
https://perfloop.ai/t/oss/case_mtwn2vqphj
Verdict
VERIFIED · settled 2026-10-21
What happened: The assertion is violated on the comparison and satisfied with this change.
Hypothesis
Make the generality board count pinned byte-search entrants that already implement a cell's complete matching language, even when its input contains non-ASCII text. This prevents favorable comparisons caused by coarse tier exclusions. It changes comparison correctness, not matcher execution, and does not establish any performance target.
Required scope is the unchanged owner-pinned200-cell board at22ed84b725fd1390ace72bccd14481e0fcbcb071, Go1.26.6/Unicode15, seed0xca5e1b0a4d and original digests. The immediate bound is thirteen missing entrant-cell obligations across seven ordinary folded cells: RustAC on length/3-4/03,script/ascii/01,case/ci/09,corpus/russian/02,size/1MiB-4MiB/00,op/indexfold/00 andop/indexfold/04; Veloz on the same singleton cells exceptscript/ascii/01. Preserve the other declared slots, all88sensitive operations and full-family equal-cell denominators. All thirteen are required coverage, not optional samples.
The supported operation is BenchmarkBoard from the arena module. It calls prepareBoardCell, then boardEntrant before oracle checks and paired timing. boardEntrant at arena/board_bench_test.go177–276 excludes folded RustAC solely when asciiTier is false and similarly excludes Veloz before considering complete language equivalence. scripts/verify_board.py77–87 repeats that coarse rule; merely emitting extra active metrics would be rejected by today's verifier. Those are the source-owned repair sites.
The equivalence is narrower than general Unicode support. When every member of every relevant fold orbit is ASCII, ASCII-caseless byte search implements the language on valid UTF-8 text. When every unit is fold-invariant, RustAC's existing exact constructor implements the same language. Veloz's ASCII-byte folding can implement certified singleton languages with invariant non-ASCII units. Casei already enumerates full fold orbits in plan.go1033–1048 and1465–1497. RustAC's existing Go/native guards and exact-mode behavior are explicit at arena/rustac/rustac.go57–82 and arena/rustac/src/lib.rs64–86. Do not infer equivalence from a corpus flag or a single negative answer.
A fresh parent audit validates112 original folded digests and617 scoped slots:604declared plus these13unchanged-API equivalents, with78positive/case/bit-five valid-UTF8 controls. Full ordinary operation streams agree with the oracle using SHA-verified original native libraries; no native library was changed. These are retained Research observations, not implemented Case proof. The first604-slot audit only repeated tier gates and is expressly not a complete supporting-field certificate.
This is a required comparison-integrity prerequisite with zero claimed matcher CPU removal. It can discharge the thirteen identified omission obligations and keep future engine comparisons honest; it cannot close board-family, public-row or final combined-source speed gaps. One mixed byte-language RustAC constructor question remains outside unchanged API equivalence and must not be silently solved by a guard bypass. This proposal adds no consumer-visible casei API, type or configuration. Internal benchmark eligibility metadata and verification tests must evolve together. The current repository actively maintains the board and field; maintainer acceptance of this particular repair is unknown.
For valid UTF-8 board inputs, whole simple-fold-language certificates can admit the thirteen demonstrated unchanged-API entrant-cell pairs without changing original IDs, consumed widths or availability policy. Mere agreement on a no-match fixture is insufficient; constructor guards outside those equivalences remain authoritative.
At accepted22ed84b with Go1.26.6/Unicode15 and the original200cells, seed and digests, establish the required roster through BenchmarkBoard/prepareBoardCell rather than only a private forced-tier wrapper. On one qualifying named Intel host with all pinned native availability gates enabled, require all thirteen demonstrated additional pairs—seven RustAC and six Veloz across the seven named folded cells—to be active, oracle-checked and represented in positive finite timing metrics. Check every other declared slot, all88exact operations and every family's unchanged denominator. Differentially test the complete language certificates and native operations with positive/prefix/tie/non-overlap/source-width controls, all orbit members, invariant non-ASCII units, bit-five near misses, and Kelvin/long-s counterexamples that must not receive an invalid certificate. Verify existing feature-availability exclusions still apply. Require matching independent verifier expectations bound to the original cell identities/digests: removing any supported entrant or corrupting its active ratio must fail, and extra unsupported entrants must fail. Retain strict largest-active-ratio,1.10cell and below1family rules. The currently losing library may still fail those performance rules; their pass is not this correctness Case's property. Reject if an admitted API mode is not equivalent, a known supported pair remains omitted, original-ID/width semantics change, source/fixture identities drift, verifier expectations merely trust reported active flags, native guards are bypassed, or availability/ratio/performance checks are weakened.
The617-slot result is a bounded current-mode certificate, not a theorem about every hypothetical underlying-library encoding or adapter transformation. The mixed count/2/01 byte language still encounters unchanged RustAC Go/native caseless guards; no altered native library, language expansion or new matching API is part of this repair. Supported benchmark inputs are the generated valid-UTF8 board population; malformed/opaque library semantics remain separate controls of matcher Cases. The repair must preserve hardware exclusions, not manufacture availability. No production telemetry, frequency estimate, measured matcher saving, field leadership or final-revision achievement is asserted. The old owner-closed Rebar reporter had a different operation and width-gated contract; it is neither resumed nor relied on here.
Change to test: Replace corpus-ASCII-only exclusions with complete-language eligibility for unchanged RustAC and Veloz API equivalents, using RustAC's existing exact mode for wholly fold-invariant literal sets. Update independently checked board-verifier expectations with the same source-bound certificates, preserving native availability, original IDs, widths and all performance rules.
Where it lives
perfloop-oss/casei · arena/board_bench_test.go
Evidence
The assertion is violated on the comparison and satisfied with this change: `On the reported GenuineIntel family 6/model 143 host with AVX2 and AVX-512F/BW/VBMI enabled, under Go 1.26.6/Unicode 15.0, BenchmarkBoard checks the original valid-UTF-8 200-cell board (seed 0xca5e1b0a4d and original digests), activates, oracle-checks, and records positive finite ratios for the 13 original RustAC/Veloz pairs across their seven full cell identities, preserves every other supporting slot, all 88 exact cells, equal-cell family denominators, original pattern IDs and Each source-byte widths, and its independent verifier binds the pairs to the original names and digests, rejects unsupported entrants and corrupt ratios, and retains the largest-active-ratio, 1.10-cell, and strict below-1.0-family rules.`
Checks: 5 of 5 passed. Verification: no defect found.