Branch tall-mountains-v1, new opt-in fracture-set keys flake*. The face is a sheet broken into 3D Voronoi flakes (elongated along the joint direction); a clustered share of flakes has fallen, removing the sheet there and exposing a floor parallel to the surface. Three sheets (0.9 m / 0.25 m / 0.07 m thick, flakes ~4 m / 1.4 m / 0.5 m across), each cut from the surface the last left, plus long vertical cracks. Test wall, grey, camera square to the 75° wall at 22 m (about 20 m across, the estimated scale of the reference photo), sun 20° off the view, 50° up. The reference photo is not shown here (it is someone else's). Click any image for full size.
Round 2: the sawtooth was the test wall's SDF (2026-10-15)
The test wall's SDF was z - height(x, y), a height difference: on the 75° wall it overstates distance 3.9×. Every depth the kernel reads was 3.9× too large: sheets came out 3.9× thinner than set (the thinnest, nominally 7 cm, was 1.8 cm, below the 2 cm mesh cell: dotted, toothed rims), and the cells were looked up at the wrong place off the surface. With a true distance (the wall's height difference divided by its slope factor, opt-in true_sdf in the lab recipe) the rims are clean. The same distortion applied to every earlier lab render on this wall (the sheeting steps' "1.6 m" were ~0.4 m), and to the rock-surface relief: here it is scaled by 0.26 to keep its earlier look (relief_scale). Kernel fixes made along the way (kept): flakes are prisms through the sheet, thickness varies smoothly, edges wander slowly, distance to a scar's edge is continuous and normalised along the face.
Diagnosis: flakes alone on the bare wall with the old SDF: aliasing everywhere (the same field sampled directly at 1 cm with a true distance was clean)v8r: true SDF; sheets 0.25 / 0.08 / 0.04 m (v7's effective thicknesses, the thinnest raised to two mesh cells)v8r_nominal: true SDF with v7's numbers as written (0.9 / 0.25 / 0.07 m): far deeper; a few combs remain on the deepest walls
photo
v7 (round 1)
v8r
v8r_nominal
edge density
0.28
0.18
0.23
0.24
relief contrast (sd)
0.24
0.17
0.20
0.25
vertical share of edges
0.50
0.46
0.46
0.43
horizontal share of edges
0.08
0.10
0.10
0.10
dark fraction
0.10
0.05
0.06
0.07
dark lines per 100 m²
7.3
7.5
10.1
8.8
dark line length, 90th pct (m)
7.0
2.6
2.8
3.5
Round 1: before and after
Measured against the photo (at the same 3 cm/px, photo assumed 20 m across)
photo
P35 (before)
flakes v2
flakes v5
flakes v7
edge density
0.28
0.04
0.18
0.17
0.18
relief contrast (sd)
0.24
0.09
0.16
0.17
0.17
vertical share of edges
0.50
0.54
0.32
0.40
0.46
horizontal share of edges
0.08
0.11
0.17
0.10
0.10
dark fraction
0.10
0.02
0.04
0.04
0.05
dark lines per 100 m²
7.3
2.6
5.3
7.0
7.5
dark line length, 90th pct (m)
7.0
6.0
4.4
3.5
2.6
The photo's figures include colour (dark water streaks, orange patches), so its contrast, dark fraction and edge density overstate its relief somewhat.
Variants
v2: first try; cracks on every flake edge read as scratches, too many horizontal edgesv5: longer flakes, cracks on large/mid edges only, deeper large flakesv7: large flakes 6x longer than wide, small flakes straighter
