integrable-lattice-verify

Project-specific verification for integrable-lattice candidates. Use when Codex needs to check whether a generated unresolved statement candidate is already solved, follows from a known theorem, is il

SKILL.md

GitHub

Integrable Lattice Verify

Goal

Reduce false candidates. Verification does not prove the candidate; it tests whether it is still a credible unresolved target.

Required Context

Read:

  • CLAUDE.md / AGENTS.md (mandatory before starting) — the repository's top-priority conventions: proof form (structured text under structured-latex/; never author new Typst), naming rules, verification commands, the definition of "done", and "文書・定理を番号や記号で管理しない" (never manage documents, theorems, or classifications by number or symbol — call them by name).
  • integrable-lattice/README.md and integrable-lattice/docs/tasks/auto-loop-runbook.md — project goal and the auto-loop procedure.
  • docs/discussion/対数順序群上の統計力学/ and docs/discussion/可算性の効用/ — primary sources for the countable core (ℕ⊂ℚ⊂Λ⊂ℚ̄) and for treating ℝ/ℂ as a convenience for applications; the countable/ℝ boundary is the decidability boundary.
  • docs/research/R-Lambda-duality/ — the ℝ-side / Λ-side duality and what counts as a closed form on the countable side.
  • docs/research/場の量子論の数学的定式化/ — which layers of existing physics close algebraically and which escape to ℝ.
  • the candidate record from integrable-lattice/outputs/candidates/
  • its known anchors and references
  • relevant seed queries and aliases for the model/operation

Workflow

  1. Search exact title-like phrases from the candidate statement.
  2. Search by aliases: model name variants, boundary condition variants, operation variants.
  3. Check whether a more general theorem implies the candidate.
  4. Check whether the candidate is ill-defined or incompatible with the model.
  5. If feasible, run or outline small-size symbolic/numeric checks.
  6. Update risk fields:
    • resolved_risk
    • novelty_risk
    • definition_risk
    • counterexample_risk
  7. Preserve links and evidence for every risk decision.

Output Shape

candidate_id:
verification_status: plausible | solved | probably_solved | ill_defined | counterexample | needs_more_search
resolved_risk:
novelty_risk:
definition_risk:
counterexample_risk:
evidence:
queries_used:
followup:

Guardrails

  • Do not call a candidate open merely because no match appears in one search.
  • Do not discard a candidate because a neighboring result exists; identify the exact implication if claiming solved.
  • Separate mathematical counterexamples from failed implementation experiments.