D-258 rulings 5 and 6: the pre-seed biome input is class ids only, on two
maps, because compute_biome was already computing them apart and throwing
one away. It gave every land pixel its Whittaker (climate) class, then painted
terrain over it (ocean bands, ice, altitude snow, mountain rock, lava and ash,
the dry, lunar and ferric ground).
planet_simulation:
- compute_biome_layers returns (biome, terrain). Each rule declares the layer
it writes, by RULE not by class: ice from the cold climate box is biome,
altitude snow is terrain with the lowland biome kept beneath. Rules that
place life (thermophiles, mats, crust, the oasis rings) write the biome and
clear the terrain under them.
- 255 means "nothing on this layer", not 0, because class 0 is ocean_deep,
itself terrain. Ids stay literal.
- compute_biome is now compose_layers() of the pair, so the renderers are
untouched. Proven byte-identical, dtype included, on 65 real bodies: every
8th standard-atmosphere body plus every thin, thick, reducing and trace one,
captured before the edit and compared after it.
- simulate() carries both maps as biome_layer / terrain_layer.
reach atlas planet bake-biome [--body --limit --dry-run --check]:
- Writes biomemap.png + terrainmap.png (8-bit, 1024x512, tEXt: layer, none,
classes, decision) beside each heightmap. Scope is the heightmap bake set
with an atmosphere (257 bodies; airless are deferred, per D-258).
- Pre-seed by construction: simulate() is keyed on the body frontmatter's
seed, and no world seed is accepted anywhere.
- --check re-simulates and compares DECODED PIXELS, never bytes, since PNG
encoding drifts across Pillow/zlib versions (T-1291) and a check that trips
on that gets muted. Mutation-proven: one flipped pixel exits 1 and names
the file; the restored file exits 0.
- import_heightmaps' body lookup is extracted as body_def_for and shared,
not copied.
Tests (make test-tooling): rock keeps its biome beneath, thin-atmosphere
ground is terrain with only scattered life, water is terrain only, no pixel is
empty on both maps, and stacking equals the rendered grid. Routing rock
through the biome map fails two of them by name. The router drift test covers
the new verb.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
compute_biome receives absolute kelvin (the Whittaker table is kelvin; the
0-1 normalisation happens afterwards in simulate(), for the renderer only),
but two terrain rules compared it to 0.35. That was the case from the
generator's first commit (f84275edf):
- mountain rock, `temperature < 0.35`: never true, so the class never
appeared through this rule on any body;
- sulfuric scrub, `temperature > 0.35`: always true.
0.35 can only mean a fraction of the world's own range, the normalisation
compute_moisture already uses, so both rules now compare against that. A
world with no temperature range reads 0 (the cold end).
tooling/test_planet_biome_rules.py pins the rule both ways on a synthetic
body: cold high ground turns to rock (>90%, since anomaly scatter repaints
~3%), and warm high ground and low ground do not. Mutation-proven: restoring
the kelvin comparison fails it at a rock share of exactly 0.000. Wired into
make test-tooling.
Sulfuric scrub remains unreachable for a different reason: it converts rock
only at elev_norm 0.25-0.65, and rock exists only above 0.65. It is left as
found, since making it reachable is a design call and belongs to the balancing
pass, so it is not pinned.
Found while shaping the D-258 biome bake. The committed reliefmaps and globes
were rendered with the bug, and the coming bake reflects the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jpmschweitzerandClaude Opus 5.5