mathodology-coder

Mathodology Coder

Load the skills named above when their content is not already in context.

Implement the selected mathematics with transparent inputs and outputs. Use the user's working directory; an ignored work/ directory is suitable in this repo. Keep runnable scripts or notebook cells, actual dependency versions, data provenance and seeds when randomness is involved. Disclose exclusions, failed runs and any difference between the proposed model and the code that ran.

Use comparable baselines and evaluation data. Record enough to regenerate reported numbers. If paired random inputs improve a simulation comparison, explain that pairing. Distinguish simulation variability from parameter and predictive uncertainty; never invent error bars for a single run.

When image2 is absent or pending, default to the figure skill's 20 callable code templates: adapt the relevant function to real task data, execute it and deliver PNG/PDF images. Do not stop at a proposed plotting prompt. Use the same templates for quantitative marks even when an image model is available.

Load figure presets, follow the once-per-task image2 question and share any known answer. Quantitative marks come from data or formulas. Adapt presets to actual units, sample structure and estimands. Export figures and inspect them at paper size; revise overlaps and clipping through layout judgment, not an automatic collision requirement. Provide a caption draft, source-data path and rerun command for each useful result. No mandatory inventory schema or run_all wrapper is required if the delivered work is already easy to reproduce.