图表证据

用途

Select, generate, or review figures that prove a claim. Main-body figures must carry mechanism, comparison, robustness, or decision evidence; diagnostic and dense support figures default to appendix.

Read math-hub/references/quality-contract.md only when promoting a figure to paper-ready, classifying blocked status, running a final gate, or writing a cross-module handoff. Local chart choice or visual repair can stay lightweight.

Figure First Principles

• One main-body figure should prove one central claim.
• Prefer tables for exact values and figures for patterns, mechanisms, tradeoffs, and robustness.
• Do not include every generated figure.
• Write a caption, one post-figure conclusion, and one boundary/risk note near the figure.
• A figure with unknown source, unit, scenario, baseline, or validation status is not paper-ready.

Evidence Contract

figure_evidence.csv should record figure id, claim id, source output, role, caption target, post_figure_conclusion, risk note, and validation_status.

The validation_status must be paper_ready before a figure supports abstract, conclusion, or final answer wording. Otherwise, keep it planned, generated, checked, blocked, or diagnostic.

A paper_ready figure_evidence.csv row must include either render_check_status=passed or human_visual_check=passed, plus a final-size readability note. A generated file is not paper-ready simply because it exists in an experiment directory.

For simulation figures, raw traces stay appendix unless they directly prove a required mechanism or failure mode. A single-run screenshot is diagnostic, not a paper-ready simulation result.

Selection And Review

1. Identify the claim the figure must prove.
2. Select the smallest chart type that makes the claim visible.
3. Verify source output and units.
4. Generate or inspect the image.
5. Check caption, conclusion, and boundary sentence.
6. Decide main-body versus appendix.
7. Return blockers to math-code, math-model, math-verifier, or math-hub.

Read references/selection-guide.md for chart selection, references/paper-figure-style.md for Nature-style palette, title, and formal-paper figure defaults, references/high-impact-chart-assets.md when local templates are visually or structurally too weak, references/visual-readiness-loop.md for visual inspection, and math-hub/references/figure-evidence-rules.md when figure evidence rows matter.

When a requested figure matches a reusable chart family, read references/chart-template-index.md, apply references/paper-figure-style.md, and adapt the closest script from assets/high-impact-templates/ after the claim, source output, unit, and evidence boundary are known. Use assets/chart-templates/paper_style.py when a claim needs a bespoke matplotlib chart rather than a vendored Nature pattern. Treat templates as starting assets, not proof; every adapted figure still needs source, caption, post_figure_conclusion, risk note, and validation status.

For main-body figures, read references/paper-figure-style.md and use the Nature-style contest defaults: calm palette, vector-first export, no in-figure title, light grid, readable labels, and a single claim sentence before plotting. Use assets/chart-templates/paper_style.py when writing matplotlib code.

Visual Readiness

Inspect saved figures before labeling them paper-ready. Block or demote when text is unreadable, labels are clipped, units are missing, legends conflict, colors obscure groups, axes mislead, marks overlap, or the figure does not answer its claim.

For final evidence rows, keep figure_id, render_check_status, human_visual_check, and visual_check_note traceable to the saved file. Use the experiment output path until the claim, caption, units, and readability checks clear; only then copy or reference it as a paper figure.

Output Contract

Return:

• selected or rejected figures;
• figure role and claim id;
• caption and post_figure_conclusion;
• render or human visual check status;
• figure_evidence.csv rows or edits;
• main-body versus appendix decision;
• blocked figures and repair owner.

Return to hub: math-hub.

Red Lines

• Do not treat attractive figures as evidence.
• Do not let a figure support a claim without source output and validation status.
• Do not use screenshots as final simulation evidence.
• Do not hide exact central results in a chart when a table is needed.