Score the rendered example
Scores apply to a specific recipe revision and its example content. They are editorial judgments. They do not measure model reliability, prove technical correctness, or substitute for a visual review in your website.
The initial scores were assigned by Codex on 7 September 2026. Each recipe records the scores, revision, review date, and a short rationale. A human calibration pass is still needed.
Visual appeal: 0–100
Assess five criteria, each worth up to 20 points:
| Criterion | What to inspect |
|---|---|
| Legibility | At the intended display size, read every essential label without zooming. |
| Hierarchy | Identify the primary reading direction and the most important element. |
| Alignment and spacing | Check shared baselines, node padding, and consistent gaps. |
| Connectors and grouping | Identify edge endpoints; check for unintended crossings and ambiguous ownership. |
| Restraint and coherence | Check that type, color, and shape choices serve the explanation. |
Modification ease: 0–100
Assess five criteria, each worth up to 20 points:
| Criterion | Editing task |
|---|---|
| Label changes | Replace the main labels with text 50% longer; inspect fit. |
| Theme changes | Switch light, dark, and paper themes; inspect contrast. |
| Structure changes | Add one node, row, message, or code line without breaking the reading order. |
| Layout changes | Change the output width or reorganize one group; inspect related coordinates. |
| Portability | Export SVG, import JSON, and confirm the content and style survive. |
For each criterion, use 0 for unusable, 5 for substantial repair, 10 for workable with manual effort, 15 for minor corrections, and 20 for a clean result. Intermediate values are allowed when the review note explains them.
Ranking
balanced = round(visual × 0.6 + ease × 0.4)
The gallery starts sorted by visual appeal. You can sort by modification ease, balanced score, or name. The rankings page shows committed balanced scores. The gallery incorporates the latest local review for each recipe when one exists.
Use the two scores separately when selecting a form. A beautiful diagram that needs extensive rerouting may be a poor starting point for a page you change weekly.
Record a review
Open Review this diagram on a recipe page. Enter the two totals, your name, and the edit or inspection you performed. Save the review to retain a dated history in your browser.
Export the JSON review history to propose a shared update. Clearing browser storage removes local reviews. There is no central review server or public voting system in this release.
For a committed update, change the recipe’s visual, ease, rationale, reviewed, and revision fields, and add the before/after evidence to docs/score-history.md. Include the old scores, the new scores, reviewer, date, and reason. Do not silently replace earlier assessments.
What the baseline scores do not claim
The first release includes rationale and totals, but not completed five-part scoring worksheets or measured editing times for each recipe. Treat the totals as provisional priorities for a human review queue. New reviews should record the criterion breakdown or the concrete editing task in their notes.