Specify a scene-referred pipeline (D19)

The spec missed Ansel, and with it Aurélien Pierre's argument that a
display-referred curve early in the pipeline throws away what every
later stage needs. Read against that argument, the code broke it four
ways: the base curve was flat past its last point and so clipped every
recovered highlight; the detail stage was handed curved, clipped values
while its comments promised linear ones; the edits ran in camera RGB
because the matrix had drifted to after them, against ARCH §5.2; and the
tone curve clamped to [0, 1] around a 2.2 gamma mid-chain.

D19 records the decision: unbounded scene-linear colour from the matrix
to one view transform, last, after the detail stage. FR-DEV-3j specifies
that view transform as an adjustable operation (contrast, white) with a
default fitted to where the retired default curve put middle grey.
FR-DEV-3e retires the per-body base curves, whose own file called them
hand-tuned shapes of unknown provenance. FR-DEV-3f makes a film stock
the view transform when one is chosen, instead of a rendering at order
25 that the later edits acted on. FR-DEV-2 gains the acceptance test
that holds the rule, and ARCH gains §6.14 and the corrected §5.2 order.

docs/dev/architecture.md

@@ -355,11 +355,12 @@ RawImage (sensor data, CPU)
 ├─────────────────────┤
 │ AI denoise          │  optional; raw-domain, joint with demosaic where possible
 ├─────────────────────┤
-│ camera profile      │  matrices + per-body base curve (FR-DEV-3e)
+│ white balance       │  camera RGB: as-shot, then the operation
 ├─────────────────────┤
-│ → working space     │  linear, wide-gamut, f16
+│ camera profile      │  the matrix (FR-DEV-3e) — no curve (D19)
+├─────────────────────┤
+│ → working space     │  linear, unbounded, f16
 ├─────────────────────┤
-│ white balance       │
 │ exposure/contrast   │
 │ highlights/shadows  │  ← masks apply per-op from here down
 │ tone curve          │
@@ -368,10 +369,11 @@ RawImage (sensor data, CPU)
 │ spot removal        │
 │ sharpen / NR        │
 │ lens corrections    │
-│ look (HaldCLUT)     │  FR-DEV-3f
 ├─────────────────────┤
 │ geometry            │  crop, straighten, rotate
 ├─────────────────────┤
+│ view transform      │  sigmoid, or the film stock (FR-DEV-3j)
+├─────────────────────┤
 │ output transform    │  → display or export profile
 └─────────────────────┘
    │
@@ -381,6 +383,14 @@ RawImage (sensor data, CPU)
 
 Working precision is f16 in a linear wide-gamut space, quantising once at the output transform.
 
+**Scene-referred until the view transform (D19, §6.14).** Everything between the matrix and the
+view transform is linear and unbounded. The view transform is the one stage allowed to compress
+the scene into a display range. With a detail stage it runs as a dispatch of its own after the
+detail passes, generated by the same composer as the fused pass so that it gets the mask layers,
+the film tables and the grain's source position. Without one it is the fused pass's tail. Both
+the view transform and the output transform (primaries, gamut clip, encode) come after
+everything that reads a neighbourhood.
+
 **A hot or dead photosite is repaired before the demosaic, not after.** Past it, one photosite of
 nonsense is a coloured cross three pixels wide that no later stage can tell from detail. The pass
 (`shaders/hot_pixels.wgsl`, run by `Demosaicer::run` into a second buffer) replaces a photosite that
@@ -1255,6 +1265,17 @@ differently, f16 rounding varies. Cache keys and graph hashes are computed over
 state, which is exactly deterministic. Cross-platform *rendering* equality is a bounded tolerance
 (R1), not a checksum.
 
+### 6.14 Scene-referred until the view transform
+
+Added 2026-09-27 (D19). Between the camera matrix and the view transform, values are scene-linear
+and unbounded, and no operation clamps above 1.0, applies a transfer function or maps to a display
+gamut. The view transform (FR-DEV-3j) is the one stage that does, and the output transform after
+it clips and encodes. This is the constraint the base curve broke and ARCH §5.2 had drawn all
+along: a display-referred curve in the middle of the chain throws away what every later stage,
+the neighbourhood ones above all, needs. A test (`scene_referred_until_the_view`) runs every point
+operation over a ramp to 16.0 so that a fragment that clips fails the build rather than the
+photograph.
+
 ---
 
 ## 13. Decisions
@@ -1278,6 +1299,7 @@ Full rationale in [requirements.md §8](requirements.md). Summary:
 | D13 | Face inference runtime and model licensing | **Runtime answered**, reopened for per-device backends (docs/inference.md); licensing open |
 | D14 | Segmentation source for local masking | Decided — arm C (docs/segmentation.md §14) |
 | D15 | Target devices — 12-inch tablet and desktop, no phone | Decided (requirements D15) |
+| D19 | Scene-referred pipeline, one view transform last | Decided (requirements D19) |
 
 ---