W-10: material diversity (presentation-only soil split + slope outcrops)

[mrtlvmbt]

Summary

W-10: Presentation-only material diversity for soil cells. Post-classify pass rewrites surface_material byte to split Soil into {SoilDry, Soil, SoilWet} based on MEASURED moisture thresholds (not folklore), and exposes bedrock outcrops for steep cells.

• Substrate MaterialId, biome cascade, caps: UNCHANGED (substrate-level tests verify byte-identity)
• Gated by any-landform-ON + enable_w10 flag (OFF path byte-identical)
• Thresholds MEASURED and pinned from Phase-0 @512x2 seeds → target 3 visible regions

Phase-0 Measurement & Threshold Calibration

Metric	Seed 1	Seed 2
Soil cells	254,475	254,280
Moisture histogram (0=dry, 1000=wet)
Deciles [p0…p100]	[3,3,3,3,7,11,23,42,78,144,1000]	[3,3,3,3,7,11,23,42,82,167,1000]
CANDIDATE-A (15/80)
SoilDry (< 15)	131,092 (51.5%)	130,097 (51.2%)
Soil (15–80)	74,621 (29.3%)	71,702 (28.2%)
SoilWet (≥ 80)	48,762 (19.2%)	52,481 (20.6%)
Slope histogram
Deciles [p0…p100]	[0,0,0,0,0,0,1,1,2,3,140]	[0,0,0,0,0,0,1,2,2,3,127]
Outcrop (slope ≥ 5)	18,585 (7.3%)	19,409 (7.6%)

Threshold Selection: CANDIDATE-A (15/80) produces ~51% / 29% / 20% distribution, which is within the target range of 40–60 / 25–35 / 15–25 for three visually distinct, coherent soil regions.

Pinned Constants:

const SOILDRY_THRESHOLD: i64 = 15;    // Captures ~51% of Soil cells (driest half)
const SOILWET_THRESHOLD: i64 = 80;    // Captures ~20% of Soil cells (wettest tail)
const OUTCROP_SLOPE_THRESHOLD: i64 = 5;  // Exposes rock on ~7% of Soil cells

Implementation

• Stage (a): Soil → {SoilDry (u8=9), Soil, SoilWet (u8=10)} by moisture within classify loop
• Stage (b): Soil* with slope ≥ 5 → Bedrock (u8=4)
• Gate: Gated by enable_w10_diversity && any_landform_on (OFF path byte-identical)

Tests (Local Results: 4/4 PASS)

cargo test: 4 passed, 421 filtered out (6 suites, 0.10s)

Tests verify:

• ON-path invariance: biome/caps byte-identical WITH-pass vs WITHOUT-pass (proves presentation-only)
• OFF-path identity: whole WorldFields byte-identical when W-10 disabled
• Presentation sanity: all surface_material bytes valid discriminants (0..10)
• Patch coherence: SoilDry (dominant class) forms coherent spatial patches (smoke test)

Acceptance Status

• ✓ compile-check: PASS
• ✓ Local tests: 4/4 PASS (all W-10 tests quoted above)
• ⏳ CI: run in progress
• Material diversity: @512 all-ON produces 3 visible Soil regions + outcrop Bedrock + 9 natural materials

Closes #438

[mrtlvmbt]

PM Authoritative Critic Verdict ✅ PASS

Robustness finding F1 fixed.

F1 [robustness] ✅ FIXED

Finding: map_dump.rs did not handle W-10 SoilDry (9) / SoilWet (10) discriminants → colour fallback corruption + histogram truncation.

Fix applied:

• Added colour cases: SoilDry (9) → darker green [64, 96, 42], SoilWet (10) → lighter green [128, 164, 90]
• Widened histogram: [0u32; 8] → [0u32; 11] (covers all discriminants 0–10)
• Updated names array: added "Water", "SoilDry", "SoilWet"
• Map previews now display all materials correctly; histogram captures 11 discriminants

Evidence: compile-check PASS + 4/4 local W-10 tests PASS (quoted in commits 48ed90f + latest)

Summary

W-10 implementation complete:

• Thresholds recalibrated from Phase-0 (15/80 from 300/700)
• Invariance test rewritten (with-pass vs without-pass proves presentation-only)
• Robustness gap in map_dump closed
• All local tests passing; CI re-running to verify 4/4 green post-fix

Ready for merge once CI green.