Mathodology Figure Presets

Mathodology Scientific Figure Presets

Choose a figure from the question the reader needs to answer. Complex figures are useful when their panels expose related evidence; complexity itself is not a quality measure. Use actual data and the user's paper language.

Agent-facing guidance

Read default figure guidance when starting figure work or briefing another agent. It is the canonical reusable instruction: select a code template, fill real data, run it and inspect the resulting PNG/PDF. Use it as adaptable guidance, not a fixed pipeline.

Before the first figure

Ask once per modeling task, unless the answer is already known:

是否有可用的 image2 模型？通过当前工具、已配置接口，还是手动生成使用？

Use the host's question mechanism and continue independent analysis while waiting. Do not repeat the question in every specialist or turn. Read image2 usage for available, manual, unavailable and pending cases. Do not claim a generic image tool is image2 or collect API keys in chat.

Default rendering route

Without image2, with a pending answer, or with only manual image2 access, use the 46 callable code templates by default. Copy the matching function into the working task, bind actual data, execute it, and export PNG/PDF. Do not stop at a plotting prompt or wait for image2. Quantitative figures use this same data-driven route even when image2 is available. Adapt the closest template when needed; never substitute synthetic preview values for missing data.

Choose and build

1. Write the figure's intended conclusion as a question before seeing the result. Identify data, units, comparison, uncertainty and the available paper space.
2. Use the base selector F01–F20 or the advanced selector F21–F46 (sensitivity screening, convergence, risk curves, schedules, routes, dynamics, model explanation, evaluation matrices). Load the relevant cards, not the entire reference collection. If the data cannot support a preset, choose its simpler alternative or explain the missing input. Prefer a figure that answers the question directly over a more elaborate one.
3. Adapt the card's plotting and caption prompts using the actual columns, measured results and scientific meaning. Do not force the data to match a preview. Follow style and export guidance.
4. Use numerical plotting tools for quantitative marks. image2 can assist illustrations, mechanism diagrams and layout concepts; rebuild quantitative layers from data. Never use generated pixels as computed evidence.
5. Inspect the rendered figure at publication size, then inspect its placement in the compiled paper. Fix illegibility and misleading encodings; remove decorative panels that do not support the argument.

When a figure reports an experiment (sensitivity, robustness, baseline or ablation comparison, repeated stochastic runs, backtesting), design it with the experiment kits: they call the task's own model, record seeds and budgets, and return inputs for the matching template and a three-line table.

Deliver the figure, a caption, its source-data or calculation location and a rerun instruction. The format is flexible. State actual limitations; do not invent statistical significance or claim that appearance implies an award.

Reference and example library

• Sources and visual references: six PLOS reference figures (including counterexamples), eight Matplotlib source-text snapshots, licenses and a per-file provenance manifest. Read snapshots before adaptation; they are not executable installation or workflow steps.
• Synthetic example gallery: previews of F01–F46 and two optional scripts, render_examples.py and render_advanced_examples.py (which also has --self-test for the advanced templates and kits). The examples illustrate presentation and statistical labeling, not evidence for a contest problem.

bashCopy code
python3 .claude/skills/mathodology-figure-presets/scripts/render_examples.py --output work/figure-examples
python3 .claude/skills/mathodology-figure-presets/scripts/render_advanced_examples.py --output work/figure-examples-advanced

The demos use NumPy and Matplotlib; the F08 template and the Sobol and profile kits additionally require SciPy. If dependencies are absent, use the host's available runtime or install them in an isolated environment; no dependencies are needed just to read and use the prompts. Existing numerical tools in R, MATLAB or another language are equally acceptable for task-specific figures.