Skip to content

Prove finite-board Snaky and explain one checked strategy branch - #18

Merged
Maverick0351a merged 12 commits into
mainfrom
codex/snaky-finite-board
Oct 20, 2026
Merged

Maverick0351a merged 12 commits into
mainfrom
codex/snaky-finite-board

Conversation

@Maverick0351a

@Maverick0351a Maverick0351a commented Oct 19, 2026 •

Copy link
Copy Markdown
Owner

What changes

The atlas can now explain one complete Snaky play while linking it to a separately checked strategy guarantee. A sibling Lean project formalizes the paper's existing result on the exact 251-cell envelope and its containing 17×17 board: Maker can force a Snaky within 21 claims from the empty position.

Finite game semantics require fresh on-board claims, count replacement claims, and require a legal Breaker reply when play continues. Immediate wins need no reply. The pinned upstream files are preserved; generated finite-card proofs use the exact 728-card certificate.

The existing explain/check interface accepts context:family-187:snaky:finite-board-21:v1. A separate adapter and independent checker save one eight-claim example. The responsive viewer shows ownership, replies, surviving envelopes, move budgets and the complete winning target. Its JSON, offline script and evidence links share one generated packet anchored to committed inputs.

Validation and CI repair

  • All 12 local proof stages passed from frozen source c3af350005f413554aea03db8cbf2c4c2d09e31b: build, kernel replay of exactly 179 local modules, allowed-axiom audit of 19 declarations, and seven mathematical rejection controls. The receipt covers 194 source inputs and 20 artifacts.
  • All 88 focused tests passed in independent WSL/Linux review, with no skips. Presentation freshness and the full renderer check pass. The 219-input publication anchor is 3bd96caa445c459371128e882fe420479084924e.
  • Independent AI review reports no open correctness findings. Earlier real-browser review covered desktop, 375/390-pixel layouts, text alternatives, keyboard controls and browser errors. The refreshed page displays the renewed receipt hash and references.
  • The prior hosted builds passed in 33–36 minutes, but kernel replay hit its 40-minute timeout. Those receipts remain failures; they did not reach the negative controls.
  • The repaired runner selects the same 179 modules through nine exact names/namespace prefixes, removing 170 repeated search-path scans. It requires the complete verbose module inventory and successful process exit. Missing, extra or duplicate modules fail both execution and saved replay. Compilation allows 60 minutes, kernel replay 120 minutes, and other stages 40 minutes. The existing required Concrete Lean certificates job allows 210 minutes and retains host capacity/resource diagnostics separately from proof evidence. Required check names and branch protection are unchanged.
  • Fresh hosted reproduction remains a release gate, reported by this PR's current checks. Local success does not substitute for it.
  • A subsequent push job exposed a test-only Linux race: the terminated child disappeared while /proc/<pid>/stat was being read. The helper now accepts the precise ProcessLookupError alongside FileNotFoundError, while still rejecting unreadable or live children. Two deterministic regressions were added; all 14 runner tests pass on Linux and Windows passes its 11 applicable tests. Independent review approved this delta; it changes no proof source, production runner or pinned receipt. Fresh hosted checks run the expanded 90-test Snaky suite.
  • All 13 immutable presentation links resolve to the expected committed files; linked project bytes match the publication anchor.

Scope and reproduction

Fresh checkout: python scripts/snaky_proof.py run --out runs/snaky-fresh, followed by python scripts/snaky_proof.py replay --out runs/snaky-fresh. Prerequisites and walkthrough: docs/SNAKY_FINITE_BOARD.md and docs/SNAKY_PLAY.md.

The renewed local run reused dependency/build caches and completed in 24 minutes 11 seconds; kernel replay took 1,303.713 seconds. This is a local measurement, not a controlled speedup study or fresh hosted timing. An earlier local bulk replay reached roughly 14.3 GiB RSS; hosted memory, disk and runtime remain reproduction concerns measured by CI.

The source paper already states the finite-board result. This work supplies its checked formalization and inspectable evidence. The displayed eight-claim play is one execution, not an improved universal bound or a formally extracted executable policy. The 616-card comparator remains a different certificate. Model benchmarks, reserved tasks and training remain paused.

@qodo-code-review

Copy link
Copy Markdown

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

@Maverick0351a
Maverick0351a merged commit 0102845 into main Oct 20, 2026
19 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant