Browser review status: passed on 2026-08-02 with Google Chrome headless at
1280px width. The source fixtures are the two reproductions from Issue #52.
“Before” was rendered from origin/main@a097c2d63eceeff9603a911a9075f0694d381f83;
“after” was rendered from this worktree.
| Reproduction | Dark | Light | Result |
|---|---|---|---|
| Dataflow default-only | before / after | before / after | After shows only data flow; PII, async, emphasis, and data-store claims are absent. |
| Lifecycle start → active → success | before / after | before / after | After shows start, active state, and terminal success; waiting and failure are absent. |
The machine-readable browser results also confirm zero
runtime exceptions/console.error calls, in-viewBox bounds, exact interactive
roles and accessible names, custom-label propagation, visual-only forced unused
kinds, long mixed-label non-overlap, and complete hidden-mode removal. The
npm run test:webm browser gate additionally exercises roving keyboard
navigation, counts and selection, canonical SVG cleanup, print/embed behavior,
guided views, and the Classic/Signal Flow/Blueprint/Editorial × dark/light
matrix. Its Dataflow fixture proves that a real database node exposes one
data store button with count 1, while the adjacent flow-variant entry stays
visual-only; Enter opens Semantic Lens on stable kind database and canonical
export strips the bridge decoration.
Checks performed in both themes:
- legend labels and swatches are visible, aligned, and inside the SVG;
- no label overlap, clipping, or unexpected empty legend region;
- node, state, transition, stage, and flow geometry is unchanged;
- no browser-visible rendering failure in either theme.
The PNGs are deterministic evidence artifacts; the generated HTML was not checked in because it is already covered by the public renderer tests and would duplicate the full viewer runtime eight times.