[yojen7]

Part of #122: phase-one print/proofing design. The issue remains open for implementation and maintainer decisions. No app processing or dependencies change.

The note describes linear Rec.709/sRGB decode, negative-primary clipping, output gamut limits, ignored non-raw ICC profiles, Linux JPEG preview and sRGB exports. It compares CPU LittleCMS proofing with a baked LUT and leaves six implementation choices to Josh. Line references explicitly identify their checked integration snapshot; links are checked.

Local review corrects the CIEDE2000 hue-wrap calculation and measures actual float16 LUT storage instead of UNORM16. The 33³ float16 simulation gives mean ΔE00 0.043, p99 0.226 and maximum 1.103 with the tested FOGRA39 profile. Those are numerical simulation results, not GPU performance or print-quality judgments. Original cloud measurements remain labelled LittleCMS 2.14; the local recheck uses 2.19.

Integrated with landed #130/#105/#132 main, preserving fixed rawler a98bd053. Hosted checks must pass on this integrated head before merge.

Source

• Original work for RapidRoom
• Harvested from another fork
• User-facing changelog entry: not applicable to this design-only change.
• Also offered upstream.

How it was made

• Written by an AI agent: original code reading/prototype/note by Claude; local integration, formula corrections, measurements and verification by Codex.
• Mostly AI-generated, reviewed and edited by a human
• Written by a human with AI help
• Written without AI

Who answers review: Yojen with Codex; maintainers decide.

Testing

• Local LittleCMS 2.19 with colord's CC0 FOGRA39L and AdobeRGB1998 profiles: all tested library proofing/output transforms agree exactly with tificc in 16-bit codes.
• Corrected CIEDE2000 passes all 34 Sharma/Wu/Dalal reference pairs within the published four-decimal rounding (maximum error 0.0000495). Tetrahedral interpolation reproduces identity at 1002 samples including endpoints.
• fmt, strict locked all-target/all-feature Clippy, 238 Rust tests (4 ignored), 112 Vitest tests, generated docs and locked release build pass.
• TypeScript: same 46 baseline errors; no changed frontend implementation, no new diagnostics, scoped Prettier passes.
• 60/60 default renders pixel-identical to approved baseline-int7 on the same Intel xe / Mesa 26.2.2 / Linux 7.2.5 environment. No baseline/tolerance change. Engine SHA-256 50f1d91eefb0bdf7fdcfcecb4297a4759d1b581b121b87a45af999f67f56462c.
• Tested by a human in the running app (no app behavior added).
• Default rendering unchanged by the full-corpus comparison.

No real paper/monitor profiling, GPU execution/timing or implementation quality claim. The output-space, library, preview path, display assumptions, first-version scope and fixture choices remain proposals. Hosted checks must pass for this exact head before merging the note.

Fresh integration logs and recovered historical session evidence retained in /home/josh/code/rapidraw-project/baseline/results/sora-final-stack/123/; temporary corpus TIFFs removed. No printer profiles added to the app.

[yojen7]

Verified and merged as c8648314 by Yojen following the maintainer queue; local integration and verification by Codex.

Print phase-one design and numerical measurement script. No application processing or dependencies changed. Six implementation choices remain for Josh; no real paper, monitor, GPU timing or print-quality validation claimed.

• 238 Rust tests pass (4 ignored), 112 Vitest tests pass. Formatting, strict locked all-target/all-feature Clippy, generated docs, scoped frontend comparisons and locked release build pass. No new TypeScript or scoped ESLint diagnostics.
• 60/60 renders are pixel-identical to approved baseline-int7, with matching environment metadata and no baseline/tolerance change. Engine SHA-256 50f1d91eefb0bdf7fdcfcecb4297a4759d1b581b121b87a45af999f67f56462c. Hosted CI is green for this exact head.
• Durable logs, comparison JSON, manifest and engine: /home/josh/code/rapidraw-project/baseline/results/sora-final-stack/123/. Successful full-corpus TIFFs were deleted after retaining evidence.