# Material trial: v4
Optional, slower CPU material, kept for comparison with older studies. New documents use watercolor-standard-v2 / paper-brush-v5 (see material-v5.md). No GPU equivalence claim; Film is a separate backend.

## Start without damaging the accepted painting
1. Save current PNG+JSON / snapshot. Read revision and scene.
2. On a detached standard scene only, set `canvas.profile="watercolor-study-v2"` and `material.model="paper-brush-v4"`. Keep aspect, dimensions, seed, resolution, roughness, absorbency unchanged. Preview; do not accept yet. Legacy/nonstandard imports are not this conversion recipe.
3. Inspect full image + crop; compare with saved baseline. Accept only the preferred result. Both version tags survive replay, export, undo and reload.

Same paper setup: output width720/grid width640, seed42, roughness.26, absorbency.42. Height follows aspect. NewDocument still accepts aspectRatio only. Example: `/sketch/skills/examples/material-v4.json` (separate trial paper, not appended to artwork).

## What differs
- Water spreads in paper coordinates; two granulation populations carry their own optical RGB mass through flow, drying, rewetting, lifting and finite reservoir brushes.
- Synthetic paper relief is averaged over each cell; visible texture does not depend on point-sampling aliasing. Paper is a neutral warm white.
- Sizing slows absorption, independently from fixation. Wet paint stays mobile longer. This is a reduced, uncalibrated model; not measured pigment species, 3D fibres, spectral mixing or the complete published models.
- Material integration is fixed at0.1s, independent of observer speed and job polling. Compute may take minutes. Submit once; poll status/job instead of retrying preview.

## Technique choices
For connected soft washes: prewet the intended brush-accessible passage, then add pigment while wet. Keep bright paper outside your strokes. Water can advance past the initially wet area: prewet is NOT a clipping mask.
For crisp boundaries: less water or a dry neighboring area; inspect immediately AND after settling. Increased wait can first spread or rim the pigment, not only dry it.
Dryness probe: branch the saved scene, append a narrow low-water stroke across the area intended for detail, then wait2s and inspect its width/edge in a crop. In a second branch, put an additional wait BEFORE the identical probe. Keep its position, width, colour and water fixed. If it still spreads, extend drying or reduce the next detail brush's water; do not infer dryness from elapsed seconds alone. Reject probe branches after learning. This is a diagnostic drawing exercise, not a calibrated moisture sensor or a universal waiting-time recipe.
For cloud volumes: a pale connected underwash can unite isolated dabs. Preserve selected whites and darker planes; less water + drying between stages can preserve structure. Broad wet hold for every layer can dissolve the silhouette.
For controlled granulation: compare the same strokes at two granulation values. V4 uses a low/high-property mixture; this is not a library of chemically identified pigments.
For pickup/softening: rewet fixed colour, let it release, then lift or use a clean finite reservoir brush. Stained pigment stays. No white erase or unlimited charge.

## Observed limits
Unchanged fast horizontal v3 strokes made strong bands in coarse-time v4. Finer time reduced bands; prewet helped more, but excess wetting caused a halo. New paper alone did not repair repeated round cloud dabs. Whole artwork and local trial outcomes differ. Do not promote a successful numerical test to an art-quality claim.

Later cottage/cloud study: an identical-stroke v4 trial looked whiter but blurred shapes. Quartering water in 24 sky/shadow strokes did not resolve it; a dry-paper version exposed flat bands. The selected result stayed v3. Treat less-water/drying advice above as hypotheses, not guaranteed repairs. See `field-notes.md` and its optional study evidence only when this mismatch occurs.

Experiment constructor overrides are not public scene fields. Unknown sizing/fixation/advection keys are rejected. Choose versioned supported tools and change painting technique; do not invent fields.