Math Modeling Full-Process Expert

Problem

Math modeling competitions often stall because analysis, coding, writing, and typesetting are handled separately. Subproblems are split inconsistently, assumptions are under-checked, results and abstracts diverge, and final papers fail with placeholders, broken image paths, or template compile errors. mathmodel-workflow-v3 turns this chain into a staged, verifiable workflow.

Workflow and Core Capabilities

It advances through plan.md and todo.md in four stages:
- Analysis and modeling: uses only explicitly numbered top-level subproblems, checks assumption sensitivity, and writes ANALYSIS_MODELING_REPORT.md.
- Coding and figures: runs Python experiments, keeps code/, results/, and figures/, and produces RESULTS_REPORT.md.
- Paper writing: supports Typst and LaTeX templates for national, Huawei, Huazhong, and MCM-style competitions; figures must come from results/ and figures/.
- Verification gate: checks entry files, section completeness, placeholders, leaked internal paths, image paths, and numeric consistency, then writes VERIFY_REPORT.md.

Boundaries

It is aimed at competition-paper delivery and does not cover DrawIO diagrams, Typst syntax tutoring, Docker, or backend deployment. For stochastic algorithms, time-series forecasting, and integer optimization, it emphasizes fixed seeds, chronological splits, constraint back-substitution, and feasibility checks after rounding.