Canvas / reference collection

Canvas parity research

Capture full pixels, text metrics and PNG bytes to compare Windows, macOS and Linux rendering.

Use Chromium 153+. Keep this tab visible. For independent references, save captures from three fresh browser processes with unchanged settings. Results stay on this page until you copy or download them.

The export includes the selected font bytes (maximum 32 MiB). Use the same file on both machines. A web font does not establish which installed face renders the family-name cases.

The reference retains the original 58-case corpus. The additional 16-case matrix varies source and destination context requests, integer/fractional placement, and nearest/high-quality scaling. It runs on the document after the reference; native CPU/GPU routing still requires tracing.

Canvas sequence test

Loading collector… If this remains visible, reload the page.

Reference export

No capture yet. Completion and repeat stability do not establish platform parity.

What is in the export?

canvas-parity-capture/v1 contains the unchanged canvas-parity-evidence/v1 reference under reference, plus a separately versioned imageMatrix (or null). Compare collector and corpus identities before comparing pixels.

RGBA, toBlob and toDataURL use independently redrawn surfaces. The matrix's 8 × 8 source control is also separate: reading the draw source first could change routing. Export methods can differ within one browser; keep their expected values separate.

This is calibration evidence, not launch-time RENDERING_PROFILE JSON. It does not identify native fonts, verify the GPU backend, or prove parity on unseen drawings. No product session is saved.