Merge PR #3402: RAW.colour: fit the per-file look on the vignetting-corrected sensor

Conflicts: crates/engine/src/camera_preview.rs: LOOK_VERSION stays 13, all three version notes kept; docs/raw-parity.md header merged (newer date, both changes)

docs/camera-preview-colour.md

@@ -129,6 +129,28 @@ still has lens corrections removed, so the pipeline applies them once through En
 Corrections. Developing this proxy requires a full-resolution demosaic; files without such
 opcodes keep their binned proxy.
 
+## Embedded vignetting correction (Fujifilm RAF)
+
+The RAF decoder turns the maker note's vignetting data into a `FixVignetteRadial` correction, and the render applies
+it (Enable Profile Corrections). The camera applied the same correction to its JPEG. Every per-file fit
+(`fit_preview`, `monochrome_look`, the near-black tone fallback) therefore takes its sensor proxy through
+`look_proxies`, which multiplies it by that gain as `fit_dng_tone` already did (`correct_vignetting`). Before
+`LOOK_VERSION` 13 the RAF fit saw the uncorrected sensor: the tone curve learned the JPEG's corrected brightness, and
+the render added the correction again. Lens-database corrections (`LIGHTCRAFT_LENSFUN_DIR`) reach the fit the same way.
+`proxies` applies the gain before it moves the sensor into the JPEG's lens geometry, so each pixel gets the gain of the
+position it samples, as in the render. The choice between the warped and the plain sensor (`alignment`) leaves out
+the pixels the warp finds no sensor for. It used to read them as black, a false edge along the border that made the
+warped candidate lose on some wide-angle shots with barrel distortion; neither change moved ΔE00 by more than 0.04 on
+the photos below.
+
+On 21 X-S10 photos the render was up to 0.3 EV brighter than the camera JPEG in the midtones (a wide-open blue-hour
+shot) and up to 0.6 EV in the darkest 3 % of the pixels (ISO 12800 indoors). On 60 other X-S10 photos, picked at
+random from one photographer's library and rendered without their sidecars, mean ΔE00 against the camera JPEG went
+from 2.71 to 2.55: 32 improved by more than 0.1, none got worse by more than 0.05. The median luminance error per
+quantile is now within ±0.04 EV, except +0.09 EV in the darkest 3 %. That rest comes from fitting on a 96 px proxy:
+each proxy pixel averages a shadow's texture before the curve, while the render applies the curve per pixel. A fit
+on a 4× finer proxy roughly halves it in a simulation, but the acceptance gates were set at 96 px.
+
 ## Camera profiles (pooled per model)
 
 One photo can show too little of a colour for its own fit to learn it: in a second shot of the lime shirt only a few dozen proxy pixels show it, next to a hillside of foliage at the same hue that the camera renders differently, and the shirt stayed yellow (hue 68° against the JPEG's 82°; a 192-pixel proxy only reached 74°). `lightcraft-cli calibrate [--max N] [--out DIR] FOLDERS…` therefore pools the colour pairs of many ARWs, NEFs or RAFs per camera model — each against its own embedded JPEG, at a 192-pixel proxy, at most 4000 pairs per photo, files spread evenly over the folders — and fits one matrix (from white-balanced camera RGB, so no single photo's white point is baked in) and one hue/saturation/value table. Models need at least 5 usable photos.

docs/raw-parity.md

@@ -1,6 +1,6 @@
 # RAW parity: cameras, formats, demosaic, colour and lens profiles
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-26 · **Change:** patch (bundled camera profile count 5 → 9) · **Target:** Adobe Lightroom Classic 15.5 / Lightroom 9.6 (Camera Raw 18.7)
+=======
 
 Raw processing is the deepest single dimension of Lightroom parity: a photographer whose camera
 does not decode, or decodes with the wrong colour, cannot switch. This document measures it model by
@@ -225,7 +225,8 @@ sample files. A pointer for contributors, not legal advice.
 - Against the camera JPEG (what the fit targets): the Sony bulk check (#535) accepted fits on 81 of 87
   bodies, median ΔE 5.1, ΔL\* +0.2; rejected fits fall back to a neutral matrix at median ΔE 18.5,
   ΔL\* −11, chroma ×0.39. RW2 140 of 174 files accepted, median 5.0. NEF 13 of 13, 2.9–7.8. RAF 23 of
-  24, pooled X-H2S 1.62 / X-T4 2.34. SRW 2.2–7.6.
+  24, pooled X-H2S 1.62 / X-T4 2.34 (measured before `LOOK_VERSION` 13 stopped applying the embedded vignetting
+  correction twice); X-S10, 60 photos not used to find that bug, mean ΔE00 2.71 → 2.55 with it fixed. SRW 2.2–7.6.
 - Against Lightroom itself (few): Sony DRO median ΔE76 7.5 → 3.3; Olympus E-1 vs Lightroom Classic
   ΔE00 11.25 → 8.30; ProRAW with its gain map ΔE00 19 (why the map is off). No measurement against
   Adobe Standard across models exists.
@@ -329,3 +330,4 @@ Beyond decoding:
 | 2026-10-14 | minor | Integration batch 36: CC0 per-model verification for 117 Nikon/Fujifilm, 97 Canon CR2/Pentax/Ricoh, 21 Sony/Leica and 23 Olympus/Samsung bodies, 22 more CR3 bodies and 30 more models; Epson ERF and Mamiya MEF decode; measured Nikon image areas (#1053, #1770): 515 verified, 675 unverified, 218 preview, 38 unsupported (re-measured with `coverage.py`) |
 | 2026-10-15 | minor | By-maker and by-use tables re-measured with `coverage.py` after batches 36–39 (545 verified, 651 unverified, 212 preview, 38 unsupported; ~94% of photographers' cameras fully work, ~85% on a verified body); rows the per-model PRs had updated are current again |
 | 2026-10-18 | minor | Minolta MRW decoded (`vendor/mrw.rs`): DiMAGE 7, 7Hi, A1, A2, A200 and Dynax / Maxxum 7D verified; 551 verified, 656 unverified, 201 preview (re-measured with `coverage.py`) |
+| 2026-10-30 | patch | RAF camera look fitted on the vignetting-corrected sensor (`LOOK_VERSION` 13): renders were up to 0.3 EV too bright in midtones and 0.6 EV in shadows; X-S10 mean ΔE00 vs the camera JPEG 2.71 → 2.55 on 60 photos |