Detect whether approved modeling code is Python or MATLAB/Beita Tianyuan and route it to the matching reviewer using the compact named-check review contract.
日本語の概要は準備中です。原文の説明を表示しています。
Audit whether the semantic evidence required by the active lean or submission profile exists and is current, without requiring one verbose artifact per skill or an arbitrary number of pass bullets.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Check deliverable evidence, not whether every invoked skill left a ceremonial report.
Per active Qx, require as applicable:
Lean completeness does not require final reports, freeze, paper, or G6 artifacts.
Per Qx require:
Globally require symbol table, assumptions, references, consistency audit, completeness audit, and QA report.
rigor_profile.paper/audits/completeness_audit.md only in submission/final mode.PRESENT: required evidence exists and is current.MISSING: required evidence absent.INSUFFICIENT: artifact exists but required semantic fields/checks are missing.STALE: cited material evidence changed after approval/review.NOT_APPLICABLE: justified by task/profile.≥5 prose pass items.PASSED only when all submission requirements are satisfied.まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Detect whether approved modeling code is Python or MATLAB/Beita Tianyuan and route it to the matching reviewer using the compact named-check review contract.
日本語の概要は準備中です。原文の説明を表示しています。
Run scoped or final cross-media consistency checks for mathematical-modeling artifacts, comparing canonical numbers, symbols, parameters, decisions, files, and paper claims without performing full-workspace audits for low-risk changes.
日本語の概要は準備中です。原文の説明を表示しています。
Map contest attachments to subquestions, audit and clean raw data, and emit one reusable data profile with quality, coverage, imbalance, concentration, and method-readiness evidence for downstream risk screening.
日本語の概要は準備中です。原文の説明を表示しています。
Build one compact choice card at a genuine mathematical-modeling judgment point. Use before method screening, after a meaningful experiment, or before final claim/freeze approval so the human chooses the trade-off while AI handles mechanical consequences.
日本語の概要は準備中です。原文の説明を表示しています。
Plan the smallest set of diagnostic, comparison, paper, and appendix figures or tables needed to support verified mathematical-modeling decisions and claims.
日本語の概要は準備中です。原文の説明を表示しています。
Build the authoritative final method explanation for a submission-ready subquestion from the method card, human decision ledger, code plan, final results, and robustness evidence.
日本語の概要は準備中です。原文の説明を表示しています。