food-figure

Food-Figure — Data-Driven Figure System for Food & Nutrition Science

Turn a dataset (or a described result) into the right submission-grade figure. The chart serves the scientific logic; polish is subordinate to making the core conclusion clear, defensible, and reviewable. Original work; architecture informed by open community figure skills (see the repo README Acknowledgements).

Load reference files as needed (progressive disclosure) — don't read them all up front. The map is in the frontmatter references list.

Workflow

flowchart TD
    A[Data or described result] --> B[1. Analyze the data<br/>scripts/analyze_data.py -> profile]
    B --> C[2. Recommend figures<br/>references/data-to-figure.md]
    C --> D[3. Figure contract<br/>references/figure-contract.md]
    D --> E{Backend?}
    E -- unknown --> Eq[Ask 'Python or R?' once<br/>scripts/backend_pref.py]
    E -- known --> F
    Eq --> F[4. Render<br/>python-guide.md OR r-guide.md + food-recipes.md]
    F --> G[5. Export at journal spec<br/>references/journal-specs.md]
    G --> H[6. QA<br/>references/qa-checklist.md]
    H --> OUT[Journal-ready SVG/PDF/TIFF + editable source]

1 — Analyze the data

If the user supplies a data file (CSV/TSV/Excel) or table, profile it first: run scripts/analyze_data.py <file> to get, per column, the type (numeric/categorical/datetime), cardinality, missingness, distribution summary, and the detected structure (grouping factors, repeated measures, time/dose axis, wide sensory/composition matrix). If the user only describes a result, elicit the same: what varies, what's measured, n, and the error type. See references/data-to-figure.md.

2 — Recommend the figure(s)

From the profile, propose the best figure type(s) with a one-line rationale each, and say what each would show. Prefer the figure that makes the paper's claim most directly; note honest alternatives. The decision rules and a full catalog are in references/data-to-figure.md and references/chart-types.md. Never force a chart the data can't support (e.g. bar-of-means where a distribution matters → box/violin + points).