[aliromman]

The complete selected P104 replay spends most of its time in literal Gaussian state evaluation. This opt-in package uses the archived exact quadratic-state engine at the two state interfaces and keeps every original scalar, native, physical-caller, invoice, compiler and outer check. On this macOS arm64 host, three alternating fresh runs per engine reduce the median from 269.00s to 98.37s (2.73×). Every full scientific stdout hash agrees. The ranges are 264.33–305.90s and 80.89–175.99s; this is a local verification speedup, not a physical multiplier speedup.

• Exact source binding preserves the selected WORD and complete RESULT, all 13 helper cuts, independent input rows, reverse columns, native phases and original rejection controls. All 13 source assertions remain explicit failures under optimized Python.
• A fresh full replay collects all 17,651 distinct state inputs. Both original literal interpreters independently match all 9,037,312 amplitude entries per interpreter, including zeros and denominators.
• The bounded cache holds immutable state signatures keyed by the complete typed input. Controls cover changed inputs, numeric aliases, mutable outputs, capacity and eviction. Three omitted proof checks and three corrupted algorithm/word/compiler inputs reject. No scientific acceptance receipt is cached.
• The committed export passes both registered commands with Python 3.12.14 under -I -O -B, empty PATH, no Git and read-only files. All exported bytes and modes stay unchanged. The normal author-side full replay, independent cross-check and six-process benchmark also pass.

Package, proof, measurements, independent trace, validation, canonical investigation.

Reproduce: python3.12 -B scripts/research_package.py --package research/selected-quadratic-replay. The two registered checks run serially and include extra literal audit work; no 2.73× speedup is claimed for that larger audit. The benchmark measures the unchanged complete selected scientific checks and is a separate six-process command in the README.

This preserves the conditional κ = .00310045 and its five assumptions. It does not change the selected record or official reproduction command. Linux timing, independent human review and proof-kernel acceptance are unrun. The checkout is shallow, so the optional full-history ancestry preflight is unavailable; all 31 source bytes and blobs, committed export and Git-free replay are checked.

Base 553d26d; frozen priority-7 card, refreshed against priority 8 before submission. Credit the exact PR5704 engine, selected PR5599 construction and all retained native/controller/compiler and classical stabilizer sources in NOTICE. The new work is the integration, controls, full-input comparison and measurements. Ali Romman with substantial OpenAI Codex assistance; no subagents. Apache-2.0. Archive publication is not independent mathematical approval.

[Clear1nsight]

CLAIM — independent Linux validation of PR5757 at 53006fe, against canonical main de5ec2e. I will preserve the complete selected RESULT/WORD, source syntax binding, absolute phase, cache controls and actual-state literal comparisons. Scope: Linux portability, complete-replay timings and qualification overhead, with serial bounded runs and explicit stop-on-drift/resource limits. No engine redesign, new construction, kappa gain or selected-frontier change. Intended result: one focused reproducible validation PR plus seven-field handoff. Target expiry 2026-10-27 02:30 UTC; I will update or release this claim. Credit Ali Romman and the retained P104/native/controller/compiler/stabilizer lineage. Prepared for Clear1nsight with OpenAI Codex assistance.

[Clear1nsight]

DONE — independent Linux qualification is published in PR #5816, exact head 5195ab6d93425bb47b2bb8f9acba05d2ae307413. It qualifies this frozen PR5757 head 53006fe6031dab5c23d5b7d1cf6a92f411a04a68 without changing the engine or source proof. Three fresh samples per timed engine give medians 339.49/130.58 seconds: 2.60x faster, with identical complete scientific output. Timed peak kernel RSS is 475.05/492.72 MiB.

Both normal and optimized full adapter checks and all-actual-state oracles pass, including every one of 17,651 distinct states against both original literal interpreters. Normal two-command qualification costs 402.88 seconds and the additional optimized qualification 404.05 seconds; these are separate from the speedup claim. Exact source bytes/modes and the original complete caller/invoice/compiler/outer checks are preserved. The package includes portable source retention, explicit resource limits, raw evidence and 17 recorded-evidence plus three actual source-entry rejection controls.

Canonical seven-field handoff. One Linux host and small descriptive timing samples; no new kappa, successor transfer, independent human or proof-kernel acceptance. Hosted CI is separate. Credit Ali Romman and the full retained predecessor lineage. Linux scope released; the kappa campaign remains paused.

[Clear1nsight]

DONE — the bounded single-pass qualification scope is published in PR #5818, exact head a0171b8d2f942cdfc5295307548fe484c570b1ba, following PR #5816. The exact subject remains this PR head 53006fe6031dab5c23d5b7d1cf6a92f411a04a68; all original source files, both entry points, complete RESULT/CROSS and both literal interpreters remain unchanged. Coverage and small feasibility passed before integration. One serial Linux comparison reduces whole qualification from 405.86 to 268.93 seconds (33.74% wall and 33.47% CPU saving), retaining all 17,651 actual inputs and 9,037,312 amplitudes per interpreter. Normal, optimized and read-only Git-free isolated optimized runs pass with identical stdout; 13 stage/contamination/exception and seven source-custody controls pass. The second complete source execution is deliberately shared, not retained as an independent cold run; the original two commands remain available. One comparison pair and one host do not establish universal performance. Seven-field handoff; exact-head hosted run is tracked separately and was running at this update. This scope is released; no new experiment or kappa research is dispatched. Credit Ali Romman and all retained engine/P104/native/controller/compiler/stabilizer lineage. New work for Chris R / Clear1nsight with OpenAI Codex assistance; no subagents.