Time RustAC in its native byte mode, not the simple-fold view, on the seven folded cells whose complete fold language it implements exactly
perfloop-oss/casei · UNCATALOGUED MECHANISM
https://perfloop.ai/t/oss/case_km0rzt46kn
Verdict
VERIFIED · settled 2026-10-22 · merged as tsenart/casei#44
What happened: The assertion is violated on the comparison and satisfied with this change.
Hypothesis
The board times RustAC through a translated simple-fold view on folded UTF-8 cells where the submitter reports that its unchanged native byte modes implement the complete fold language exactly. This may make the competitor comparison favor casei. The complete-language certificates and the timing effect on the seven submitted cells remain independently unverified. Comparison honesty only; no casei speed claim.
The submitter reports that since tsenart/casei#42, boardEntrant sends every folded non-ASCII cell for RustAC through RustACView, which rebuilds the canonical haystack and a per-byte origin map inside each timed call. At revision 0fb894708cdc77421001f1540ebeaf896bffce32, the current fallback gate is visible in arena/board_bench_test.go:371-386. RustACView owns its canonical text and origin map per operation, and Find and Each build them before native search (arena/board/rustac.go:10-13,61-89,95-133). BenchmarkBoard prepares entrants through prepareBoardCell and then pairs their bound operations for timing (arena/board_bench_test.go:37-76,149-197). This is a claim about that supported benchmark operation, not production search traffic.
The submitter's required seven-cell partition is exact mode on the two fold-invariant pattern sets, length/3-4/03 (亚) and size/1MiB-4MiB/00 (丹4); ASCII-caseless mode on the five sets whose every rune has an all-ASCII simple-fold orbit, case/ci/09, corpus/russian/02, op/indexfold/00, op/indexfold/04 and script/ascii/01. The two modes are one selector decision from one certificate, not two repairs: the orbit walk that proves ASCII-caseless equivalence is the same walk that proves invariance, and the Go test and verifier pin one seven-cell partition. These seven UTF-8 cell identities and the mixed count/2/01 control are pinned in arena/board/cells.txt:9,43,61,93,128,172,196,200; their claimed pattern certificates still need checking against the generated patterns, not just the planted matches.
A cell's x_vs_best is casei's time over the fastest supporting entrant's time (scripts/verify_board.py:17-22,170-181). The submitter reports that timing the slower view where the native mode is valid understates RustAC and flatters casei on those cells; no paired direct-versus-view timing was measured in this run. The current Go mode test and expected_rustac_modes require the view on all folded UTF-8 cells (arena/board_bench_test.go:636-694; scripts/verify_board.py:125-129). RustAC already exposes both native compile modes without a new API: exact accepts arbitrary pattern bytes, while ASCII-caseless rejects non-ASCII pattern bytes (arena/rustac/rustac.go:57-107).
The submitter reports that the certificate functions (asciiFoldOrbit, rustacASCIICaselessEquivalent, rustacExactEquivalent) and their orbit, counterexample and ordering tests already exist in closed pull request tsenart/casei#43, which was superseded only because the view now covers the roster. That external history has not been independently verified here. The submitter identifies Veloz's direct route as a separate admitted Case (case_qykpnd7r9s); this idea covers RustAC only.
The Case must check the complete simple-fold language, meaning every rune reachable by repeated unicode.SimpleFold, rather than infer equivalence from one haystack's matching answers. Run the Go mode, orbit, counterexample and ordering tests on the digest-pinned board patterns and python3 scripts/verify_board_test.py. The required signal is exactly the seven submitted folded UTF-8 cells certified with their stated native modes, reported as direct through rustac_direct_active, with no other folded UTF-8 certification. Kelvin sign, long s and mixed count/2/01 must stay on the view. Check native results against the board oracle, including original pattern-ID ordering and source spans. A nontrivial fold orbit in an exact-certified set, a non-ASCII orbit member in an ASCII-caseless-certified set, a different cell partition, or a result/order mismatch would disprove the submitted certificate. Paired BenchmarkBoard runs on the unchanged pinned cells can quantify the unmeasured timing effect without turning it into a casei speed claim.
The current field policy requires semantic agreement and includes any necessary adaptation in entrant timing (arena/field.yaml:21-32). The proposed direction changes the internal board selector, its mode tests and the verifier's expected partition, while retaining existing native modes and metric names. It adds no public library operations or configuration and does not widen other arena consumers' contracts (arena/field.yaml:21-23,75-86).
Change to test: One mechanism: a byte-language certificate in boardEntrant's rustac case, before the view fallback, from one orbit walk over every pattern rune. Every rune fold-invariant selects RustAC's exact mode; otherwise every rune's whole simple-fold orbit ASCII selects its ASCII-caseless mode; anything else keeps the view. Report the native mode as "direct" through the existing rustac_direct_active metric. Pin the seven certified cells and modes in the Go mode tests and in verify_board.py's expected_rustac_modes, with a test that no other folded UTF-8 cell is certified and controls for Kelvin sign, long s and the mixed count/2/01 cell, which must stay on the view.
Where it lives
perfloop-oss/casei · arena/board_bench_test.go
Evidence
The assertion is violated on the comparison and satisfied with this change: `For Go 1.26.6 (Unicode 15.0.0), the 77 digest-pinned folded UTF-8 BenchmarkBoard cells with nonempty valid UTF-8 patterns and haystacks bind RustAC to native exact bytes on length/3-4/03 and size/1MiB-4MiB/00, native ASCII-caseless bytes on case/ci/09, corpus/russian/02, op/indexfold/00, op/indexfold/04 and script/ascii/01, and the simple-fold view on the other 70; selection classifies every pattern rune's complete unicode.SimpleFold orbit, the seven native bindings agree with board.Matches on their pinned operations and every full-orbit pattern spelling while preserving original pattern IDs and complete Each source-byte spans, BenchmarkBoard maps the selected direct/view mode to rustac_direct_active/rustac_view_active, and Kelvin-sign, long-s, and mixed count/2/01 controls remain on the view.`
Checks: 3 of 3 passed. Verification: no defect found.