Run the automatic pipeline to generate a full competition math modeling paper from a problem statement within a tight deadline.
Use the manual checkpoint mode to step through each quality gate and keep control over the modeling and writing process.
Generate reproducible figures and Word-formatted equations where every number traces back to actual script output.
Validate a competition solution through six quality checkpoints to avoid common submission mistakes before the deadline.
Requires a Codex-compatible AI coding assistant environment and Python. The 60+ scripts cover the full pipeline from analysis to Word document generation.
This project is an AI-assisted workflow for competing in mathematical modeling competitions, specifically the Chinese national contest (CUMCM), the May Day competition (51MCM), and the American MCM/ICM competition. Mathematical modeling contests give teams a real-world problem and ask them to build a mathematical model to analyze or solve it, then write a full academic paper explaining their approach, all within a tight deadline. The system is built as a skill for Codex, an AI coding assistant. Once installed, a user can describe their competition problem in plain language and the agent will walk through the full process: reading the problem, breaking it into sub-questions, checking the quality of provided data, suggesting candidate mathematical approaches, generating code to run the models, producing charts, and finally drafting the competition paper as an editable Word document. A key design choice is the quality gate system. The workflow has six named checkpoints (G1 through G6) that must pass before the process advances. These gates check things like whether the problem has been properly understood, whether proposed methods have been validated with a small proof-of-concept, whether all numbers in the paper can be traced back to actual script output, and whether an academic integrity check has passed. This prevents common competition failures where teams rush into complex models without properly reading the problem, or where final paper numbers quietly drift from what the code actually produced. The project supports two modes: a fully automatic pipeline for tight deadlines, and a manual mode with explicit checkpoints at each stage for teams who want to stay in control. Papers must contain at least 9,000 substantive characters, and mathematical formulas are rendered as native Word equations rather than images. The project is written in Python and includes over 60 scripts covering pipeline management, document generation, figure plotting, model repair, and quality auditing.
← xuec699-sudo on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.