R-12 + glacial relief-fix: terragen preview + Option-A moraine (NOT for master merge)

[mrtlvmbt]

What

Terragen diverse-relief render preview (512²) + the plan-consensus'd Option-A glacial relief fix. Roadmap-on-branch — do NOT merge to master (per owner).

Render preview (initial)

• u16-safe batching (macroquad quad_gl overflow at 512²), adaptive rows_per_chunk, material coloring (not biome), HEIGHT_SCALE 0.4→0.2.

Glacial relief fix (Option A — 17-fork plan-consensus)

Root (Phase-0 measured): glacial deposit_till dumped AREA-scale excavated mass (~627–859 units/cell, 3–4× the [0,hmax] range) onto a 1-cell margin → clamp rim (the "ceiling plateau"); ice_mask needles (459–583). Excavated ≫ in-range capacity ⇒ un-placeable by any geometry.

• Deep troughs kept — incision/K_ICE/flatten UNCHANGED (Route-1 excavation-reduction DROPPED).
• Thin margin-peaked moraine via one run_glacial_with(…,&PROFILE,k_band) seam (budget-drain), PROFILE{d_peak=40}, K_BAND=5; hard cap ≤hmax−1 ⇒ plateau impossible by construction.
• Route-2 off-map outwash carries the large remainder (exported_till ~91–99% — faithful glacial physics; the ≤1% ε was withdrawn as physically wrong).
• P2 hole-fill S_max=16 de-needles (needles → 0 at 512).
• Erosion tail untouched; :445 ON-path clamp dropped (tail matches OFF path, F21/F22).

Acceptance

• Phase-0b GATE @dim=512 (production 11111, both seeds): truncated==0 (no plateau), needles==0 (was 459–583), median_resource≥1, solid_frac∈[0.15,0.50], ledger identity excavated==deposited+exported_till exact.
• Regression teeth in CI: (i) conservation identity; (ii) hard exported_till==0 on dim64/s2/mask00010; (iii) non-trivial moraine (deposited>0 ∧ Till local-max); (iv) deposit-referenced plateau tooth truncated==0 + positive control PROFILE.scaled(4)⇒truncated>0 (K_CTRL=4).
• F17 same-commit: ~5 till-location tests restated to true contracts (valid-range, non-ice-outside-apron, U-trough wall-drop on pre-deposit field, till-local-max); W-SIM-6: glacial relief — ELA-gated ice incision + till moraine #416 ТЗ amendment noted.
• Throwaway deleted (phase0/sweep/_scaled/old deposit_thin_band/height_stats bin); run_glacial_with promoted to production.
• Golden re-pin: only golden_vector_matches_pinned_glacial_fixture moved (index 0: →77; CI-measured, arch-agree); all other gen goldens + sim v2_golden_conserved_* byte-identical (F21).
• CI GREEN on HEAD (5ba182b, run #29235149701, all 4 jobs).
• code-critic VERDICT: PASS (3 passes; see comment).

Plan + Phase-0b numbers: animata-pm reports/terragen-height-contract-plan.md (+ PHASE0*).

🤖 Generated with Claude Code

https://claude.ai/code/session_01KSFzwj8cXU3wLePgNZWAra

[mrtlvmbt]

Pre-merge self-review (code-critic, PM-authoritative) — VERDICT: PASS

Cold code-critic fork on git diff main...HEAD vs the plan-consensus'd ТЗ (.claude/plans/terragen-height-contract.md). Ran 3 passes (initial + 2 re-verify after fixes):

• Pass 1 → FAIL: F1 (dead deposit_till, style) + F2 (wall-drop test measured POST-deposit, robustness).
• Pass 2 → FAIL: F1 fixed; F2 residual — test re-derived incised_height with a base-mask interior vs production's filled-mask interior (CI could NOT catch this — assert passed on the near-right field).
• Pass 3 → PASS: margin/interior now both from filled_mask, matching production exactly; wall-drop on pre-deposit incised_height; both U-trough asserts intact (on_spread<=off_spread ∧ on_wall>=off_wall); ≥3-segment guard present; production unchanged; determinism held.

Final: VERDICT: PASS. CI GREEN on HEAD 5ba182b (run #29235149701, all 4 jobs). This branch is NOT for master merge (roadmap-on-branch, per owner).