visual-render-qa

原文涵盖的主题

• Visual Render QA
• Purpose
• When to invoke
• Inputs
• Required intermediate representation
• Procedure
• Hard constraints
• Strong heuristics

A specification that is never viewed is never verified. This skill looks at the real output — not the intent, not the code — at the exact final dimensions and medium, and refuses to pass artifacts that lie, that cannot be read, or that fail accessibility. It is renderer-independent: the same checks apply to plotting-library output, a slide, an SVG, a PDF figure, or a screenshot. Last, before declaring any visual done, and again after every repair. Also when auditing an existing figure you did not build. The rendered artifact at final size/medium (image or the artifact itself). The intent record, encoding table, and composition/fidelity plans (to check the render against what was intended). A QA report: 1. Render and view at final size in the real medium (MEDIA-02, QA-08). For a deck, view at projected scale; for print, at print size. Do not judge from a zoomed-in editor view. 2. Apply hard gates first (QA-01, QA-02). Reject before any aesthetic scoring if any hard constraint is violated: Truncated bar baseline (ENC-02); distorted magnitude / hidden truncation / 3-D quantity (SCI-02); missing color bar on a color scale (ANNO-04); missing axis units (ANNO-05); missing data source (AN

这里只展示节选，请以完整原文和当前要求为准。