Verdict: flaky pixel-comparison tests in js-packages/profiler-layout, not caused by any particular PR. Seen dropping an unrelated, docs-only PR (#7492) from the merge queue.

Failing job: Web Console Unit Tests (run 41244625987)

Failing tests (both mount a diagram in a real headless Chromium via mountDiagram.ts and assert exact pixel values):

Why this isn't PR #7492's fault: that PR only touches docs.feldera.com/** and js-packages/web-console/screenshots/** (a different package, different test suite) — diff. It never touches js-packages/profiler-layout.

Reproduction: ran both test files 3x locally with the same Chromium build (npx vitest run --project browser diagramShadows.test.ts diagramText.test.ts) — all 9 tests passed every time. The exact pixel thresholds these tests check (alpha channel, grayscale "ink strength") are sensitive to font/anti-aliasing differences between CI runner images, which points to environment-dependent rendering rather than a logic bug.

What to look at next:

• Whether the CI runner image changed recently (font packages, Chromium version) around 2026-10-31.
• Whether mountDiagram.ts/nodeShadow.ts pixel thresholds (diagramShadows.test.ts:20, diagramText.test.ts:53) have enough margin for anti-aliasing variance across environments.
• These tests were added recently ([profiler] Mount the diagram in a browser to test the pixels it paints, 2026-08-25); no prior occurrences found in open issues.

Next step for #7492: re-queueing should pass; the failure is unrelated to that PR's changes.