This skill builds on 2analysis-modeling. The goal is to turn the models and algorithms in reports/ANALYSIS_MODELING_REPORT.md into reproducible programs, produce trustworthy results, and generate the data-based charts needed in the paper.
Mathematical Modeling Standards Reference
If domain judgment is needed, read the sections "Problem-Type Error-Prevention Quick Reference", "Code Implementation and Results", "Common Errors in the Coding Stage", and "Charts and Visualization" in ../_references/math_modeling_norms.md . This file serves only as a standards knowledge base and does not add fixed deliverables for this stage.
Stage Boundaries
- This stage is responsible for: code, experiment runs, results, results tables, data-driven charts.
- This stage is not responsible for: technical roadmaps, algorithm flowcharts, system architecture diagrams, conceptual schematic diagrams. These are handled by
4drawio。 - This stage does not write the main body of the paper; it only provides trustworthy numerical values and chart assets for
5writing.
Step 1: Code Structure
According to plan.md in "Project Directory Structure", create the code/ and figures/ skeletons, then start writing code. The number of sub-problems is not necessarily 3; adjust according to the actual number in the competition problem.
Step 2: Implement Sub-Problem by Sub-Problem
Implement in sub-problem order; do not write everything at once without running it.
Each sub-problem must complete:
- Read the required data.
- Implement the model or algorithm.
- Verify constraints.
- Output the core results.
- Draw rich charts.
- In
reports/RESULTS_REPORT.md, clearly write the method, key numerical values, and verification results.
For optimization-type problems, a feasible solution must first be guaranteed, then the objective value optimized. For prediction-type problems, a training/validation split or reasonable error evaluation must be done. For evaluation-type problems, the indicator direction, normalization method, and weight source must be explained.
Step 3: Result File Format
During implementation, solving, and plotting, the AI must save key intermediate processes as data and keep good records, such as cleaned data summaries, model parameters, iteration history, constraint checks, sensitivity analysis processes, data used for charts, and run logs. Intermediate data should preferably be saved to figures/ or code/outputs/, and the purpose of the files should be explained in reports/RESULTS_REPORT.md .
reports/RESULTS_REPORT.md Recommended structure:
# 计算结果
## 运行环境
## 数据读取与预处理
## 问题一结果
## 问题二结果
## 问题三结果
## 灵敏度分析
## 约束与一致性校验
## 与建模报告的一致性说明
## 可复现运行方式
All data and chart results must appear cited in reports/RESULTS_REPORT.md .
Step 4: Generate Data-Driven Charts
According to reports/ANALYSIS_MODELING_REPORT.md and reports/RESULTS_REPORT.md , plan the charts and generate PDFs to figures/。
Typical charts:
- Prediction type: actual value-predicted value comparison, error distribution, metric comparison.
- Optimization type: convergence curves, cost comparison, resource utilization, before-and-after scheme comparison.
- Evaluation type: comprehensive score ranking, radar chart, heatmap, sensitivity curve.
- Data understanding: distribution plots, trend plots, correlation plots, box plots.
Chart requirements:
- PDF vector output, suitable for the paper.
- Do not write a large title inside the figure; leave the title to the paper caption (Typst's
caption:or LaTeX's\caption{}）。 - Chinese papers use Chinese axis labels and legends; English papers use English.
- Do not generate flowcharts/architecture diagrams/roadmaps.
Charts may be generated by the main program or by a standalone script; a fixed script name is not mandatory. Whichever approach is used, the data source and generation record corresponding to the chart must be saved.