zchary1106/figurecraft/figurecraft-visual-critic

FigureCraft Visual Critic

Review semantics before aesthetics. Never “improve” a figure by changing data, topology, dimensions, labels, or architecture facts.

Review order

1. Data and logic: values, units, intervals, nodes, edges, directions.
2. Representation: correct chart/diagram family and reader-facing view.
3. Completeness: labels, legends, dimensions, stage counts, dependencies.
4. Geometry: clipping, overlap, edge crossing, label collision, alignment.
5. Typography and accessibility: final-size text, contrast, color redundancy,

SVG title/description.

1. Reproducibility: FigureSpec, manifest, input hash, versions, editable output.

Workflow

1. Read source, manifest, and rendered output.
2. Locate sibling figurecraft/scripts/figure.py and run lint.
3. Inspect the image at actual target scale, not only zoomed in.

Use effective final-size font metadata, not SVG coordinate font size. Reject a paper-readiness claim for an unspecified medium or unreadable resized text.

1. Report concrete, locatable failures.
2. Apply at most two targeted revision rounds.
3. Re-render and reject completion if a blocking issue remains.

Check visible provenance and whether the chosen view answers one reader question. For revisions, verify promised locks and the authoritative source. Deliver one current index with history rather than a collection of indistinguishable previews.

Data/topology errors, clipped content, overlap, unreadable text, undefined uncertainty, invalid SVG/Draw.io, and missing source are blocking.