<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> This project provides a structured workflow orchestrator for math modeling competitions, designed to mitigate common pitfalls like misinterpreting problems or unproducible numbers. It targets students and researchers by enforcing rigorous checks and traceability throughout the modeling process, ensuring reproducibility and preventing workflow errors. The core benefit is enhancing submission reliability while keeping critical modeling judgments with the user.

How It Works

The system implements a multi-stage, gate-driven pipeline (Groundwork, Method Validation, Code & Review, Results/Freeze, Paper Writing/Audits). Its novelty lies in explicit pass conditions for each "gate," ensuring no step is skipped. Key design choices include:

• Frozen Numbers: All paper data must originate from frozen_numbers.json, with modifications logged and re-frozen.
• Auditable Artifacts: Code reviews and audits are generated as disk files with specific checks and citations.
• User-Controlled Gates: Critical decisions (method selection, result interpretation) are deferred to the user, with gates failing on empty or AI-generated rationales. This approach prioritizes AI-managed mechanical correctness and reproducibility, while reserving modeling judgment for the human user.

Quick Start & Requirements

The recommended installation is cloning the repository into the contest project directory. Global installations for Claude Code or Codex are also supported.

• Primary Install: Clone into project folder: git clone https://github.com/zhnnky329/MathModeling-skills.git .skills-tmp then use mv commands to integrate .claude, .codex, CLAUDE.md, AGENTS.md, and skills-docs.
• Prerequisites: Requires Claude Code or Codex environment; CLAUDE.md (for Claude Code) or AGENTS.md (for Codex) must be present in the contest project.
• Links:
  • Repository: https://github.com/zhnnky329/MathModeling-skills
  • Documentation: CLAUDE.md, AGENTS.md

Highlighted Details

• Staged Workflow with Gates: Enforces a rigorous, multi-stage process with explicit pass conditions, preventing common workflow errors.
• Reproducibility via Frozen Numbers: All numerical results must originate from frozen_numbers.json, with changes logged and re-frozen (Gate G4).
• Auditable Code & Reviews: Code reviews and audit reports are stored as disk files with specific checks and line-level citations (Gate G3, G6).
• User-Centric Decision Making: Key modeling judgments are deferred to the user, with gates failing on empty or AI-generated rationales (Gates G2.5, G4.5).

Maintenance & Community

Bug reports, ideas, and feedback are welcomed via email to zjzhang0424@gmail.com. GitHub Issues and Pull Requests are also accepted.

Licensing & Compatibility

Released under the permissive MIT License, generally allowing commercial use. However, users must independently verify contest-specific AI-use policies, as some competitions have strict originality requirements.

Limitations & Caveats

This tool automates workflow checks and reproducibility, not core modeling or writing. Critical modeling decisions and justifications remain the user's responsibility; the system blocks submission if these are incomplete or AI-generated. Users must independently confirm their specific contest's AI-use policies, as rules vary significantly.