[akendrick]

ChartThrob-style curve workflow for process calibration.

• Export: process curves as Photoshop Curves (.acv, up to 16 natural-spline points placed greedily where the spline strays most from the unrounded curve), Photoshop arbitrary map (.amp, exact at 8 bits), a 33³ .cube LUT and CSV. App: "Export Curve…" in Process Calibration. CLI: openquad curve CURVE OUT, plus --image PHOTO with a .tif output for the curved photo.
• Smoothing: a local straight-line (tricube, LOESS-style) fit over neighbouring patches before inverting, then a least-squares never-falling fit (pool adjacent violators) in place of the running maximum. None/Light/Medium/Strong = 0/3/5/9 patches; Medium by default, kept in run.json, chart --smoothing. On simulated noisy charts Medium cuts curve roughness about 20–30× and halves the worst error.
• Lost tones: counts patches printing as paper white or full black beyond what the target expects; 3% or more gives a warning with what to change (exposure or blocking colour). Shown in the wizard's result and log and in openquad chart; stored in the round's results.
• Swift core mirrors all of it, bit for bit against new vectors (builds.json).
• Tests: pytest 466 (98% coverage), Swift core 70, app 34.

The .acv/.amp layouts follow the Photoshop file formats but haven't been opened in Photoshop yet.

🤖 Generated with Claude Code