Repository navigation
Conversation
… full combined verification
M6LI
changed the title
… response28 Withdraw the incompatible GL probability transfer, retain delta=1e-16, and change only the Gaussian fallback invoice to146478. Preserve historical evidence, exact full supplier closure, four reviewed normal/optimized assemblies and docs-only reassemblies. Include required baseline OOM failures; no current frontier record or unconditional theorem claim. Integrate PR473 exact response28 source and validation ancestry3dc7cc1. Prepared and independently checked with substantial OpenAI Codex AI assistance.
M6LI
changed the title
|
Thank you @M6LI. This contribution is now retained on main in the scoped 100-PR review and contributor credits. Contribution: Count-only Gaussian fallback correction and probability-law withdrawal. Review and limits: The revised construction retains the original conditional 10^-16 bad fraction. Earlier stronger probability claims and receipts are explicitly withdrawn; do not silently reinstate them. The complete original source, notices and relevant execution evidence are retained through the pinned source archive. The review distinguishes fresh local runs, source-bound historical Linux CI and prior exact-head review; it does not claim 100 fresh full mathematical audits. Closing this historical PR as reconciled. This preserves parallel and cumulative credit without replacing the selected frontier or promoting priced research targets to completed constructions. Reviewed head: |
4 participants
This revision reduces only the complete Gaussian fallback invoice from 194688 to 146478, retaining the original conditional orthogonal-cover bad fraction 10^-16. It adds a source-pinned response28 composition and preserves all original rows, physical programs, prefixes, coefficient charges, complex supplier data and outer assumptions.
The fallback count is 24m²+6m−6 at m=78, from the retained unit-pivot local-ring Gaussian algorithm. No stronger cover probability is inferred. Both exact roots, six bootstrap steps and all 47 strict outer constraints were independently reconstructed, with old-count-at-new-root, adjacent-grid and parameter controls.
Validation: fresh full original supplier replays completed in 1378.1s (replica84) and 729.8s (response28). Four corrected normal/optimized assemblies explicitly reused those completed full parents; four further assemblies bind the final documentation-only revision. Each pair has identical RESULT bytes. Separate reviews bind source Git blobs, complete invoices, actual programs, canonical complex payloads and terminal resource receipts. The default verifier rebuilds the original suppliers; completed-parent reuse has a distinct receipt.
Current evidence and independent review:
Required baseline checks did not pass within the declared 1 CPU/2 GB cap: make verify hit the inherited h57 producer's OOM; a fresh full gen4 run with exact SymPy1.14 passed five stages before OOM entering banks. Source gates and resource limits were retained. These failures and partial outputs are preserved separately from the successful contribution-specific replays. The exact-head workflow is provided; no future CI-success claim is made.
Correction history: the earlier GL-derived 2m³/q probability transfer is withdrawn because its sampling model did not match the retained O-cover row invoice. Historical receipts remain explicitly labelled withdrawn; the present revision keeps the original 10^-16 law conditional. This is an improvement over the two exact pinned sources, not a current frontier record: newer PR493 has a stronger source-carrier pairing whose complex component now binds. No unconditional theorem or new all-size cover/compiler/hosting proof is claimed.
Builds on PR465, PR473 and PR478. Original source ancestry, Apache-2.0 notices and attribution are preserved. Substantial OpenAI Codex AI assistance supplied the construction and separate automated reviews; human review is not claimed.