Files
codex-mmo/evals/visual-conformance
2026-08-24 08:11:59 -07:00
..
😏
2026-08-24 08:11:59 -07:00
😏
2026-08-24 08:11:59 -07:00
😏
2026-08-24 08:11:59 -07:00
😏
2026-08-24 08:11:59 -07:00

Visual-conformance evaluation

This suite evaluates visual-engineering through the complete reference-to-render path. The reference PNG is attached to the root evaluation run. The visual analyst must inspect it, the text-only implementer receives textual criteria, and the visual verifier must receive both the original and the generated actual.png.

render_preview.py is a deterministic standard-library preview renderer, so the fixture needs no browser or third-party package. It is intentionally a coarse rendering oracle: semantic HTML, responsive CSS, accessibility, and visual judgment remain separate validation responsibilities.

Reference asset prompt used with the image-generation skill: “Create a polished dark-mode SaaS inventory dashboard at desktop resolution, with a left navigation rail, page header, three KPI cards, a blue inventory trend line chart, and a recent activity panel; crisp product UI, restrained navy palette, electric-blue accent, no logos.”