jpmschweitzer/settled-reach
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

test(simulation): D-239 tile-derivation verification harness (T-1031) #165

Closed
jpmschweitzer wants to merge 0 commits from verification-harness into main
pull from: verification-harness
merge into: :main
Conversation 1 Commits - Files Changed -
jpmschweitzer commented 2026-06-08 19:39:52 +02:00
Owner
Copy Link
Copy Source

T-1031 — D-239 Tile-Derivation Verification Harness

Capstone test for the T-974 tile-derivation epic. New server/tests/derivation_harness.rs (29 tests) plus its seed→tile golden fixture, covering four domains:

  1. Golden-seed determinism regression (D-239 §1 / D-010) — pins (seed, region, chunk_pos, tile_pos) → VoxelColumn as a JSON golden; any drift in the derivation chain trips it.
  2. §8 binding-law sweeps (D-239 §8) — drainage monotonicity, lithology→landform, glaciation→form (FjordWall only at GlaciationGrade ≥ 2), climate→vegetation no-skip + riparian band.
  3. Per-family <5 ms/chunk budget (D-239 §10) — derives a full 64×64 chunk per family; hard gate under BUDGET_ASSERT=1.
  4. Validation bodies (D-239 §1) — lore-anchored, DB-free: Kallast/Glødberg/Marevna validated; Velen/Gruenfeld skipped (not in wiki) with logged reason.

D-240 alignment

This branch was rebased onto the merged T-1033 class-envelope temperature work:

  • Validation-body fixtures no longer set the removed BodyParams orbit/star fields (orbital_period_days, axial_tilt_deg, spectral_class, star_type).
  • Marevna's temperature note/assertion rewritten for the envelope model: oceanic band [-12, 28]°C + maritime factor 0.6 + standard-atmosphere greenhouse → ~+26°C at the equator (replacing the stale ~-56°C orbit-equilibrium note).
  • Harness fmt'd + one clippy lint fixed (the file predated the gate's cargo-test/clippy coverage).

Closes T-1031. Completes epic T-974.

🤖 Generated with Claude Code

## T-1031 — D-239 Tile-Derivation Verification Harness Capstone test for the **T-974** tile-derivation epic. New `server/tests/derivation_harness.rs` (29 tests) plus its seed→tile golden fixture, covering four domains: 1. **Golden-seed determinism regression** (D-239 §1 / D-010) — pins `(seed, region, chunk_pos, tile_pos) → VoxelColumn` as a JSON golden; any drift in the derivation chain trips it. 2. **§8 binding-law sweeps** (D-239 §8) — drainage monotonicity, lithology→landform, glaciation→form (FjordWall only at GlaciationGrade ≥ 2), climate→vegetation no-skip + riparian band. 3. **Per-family <5 ms/chunk budget** (D-239 §10) — derives a full 64×64 chunk per family; hard gate under `BUDGET_ASSERT=1`. 4. **Validation bodies** (D-239 §1) — lore-anchored, DB-free: Kallast/Glødberg/Marevna validated; Velen/Gruenfeld skipped (not in wiki) with logged reason. ### D-240 alignment This branch was rebased onto the merged **T-1033** class-envelope temperature work: - Validation-body fixtures no longer set the removed `BodyParams` orbit/star fields (`orbital_period_days`, `axial_tilt_deg`, `spectral_class`, `star_type`). - Marevna's temperature note/assertion rewritten for the envelope model: oceanic band `[-12, 28]°C` + maritime factor `0.6` + standard-atmosphere greenhouse → **~+26°C at the equator** (replacing the stale ~-56°C orbit-equilibrium note). - Harness `fmt`'d + one clippy lint fixed (the file predated the gate's cargo-test/clippy coverage). Closes **T-1031**. Completes epic **T-974**. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
jpmschweitzer added 3 commits 2026-06-08 19:39:52 +02:00
test(simulation): believability + budget verification harness (T-1031) a943147843 
D-239 §1/§8/§9/§10 verification harness (server/tests/derivation_harness.rs):
golden-seed determinism regression, §8 binding-law sweeps (drainage monotonicity,
lithology->landform, glaciation->form no-glacial-U-at-grade-0, climate->vegetation
no-skip + riparian), per-family <5ms/chunk budget assertion, and validation-case
bodies (Kallast=alluvial pass; Velen/Gruenfeld skipped-not-in-wiki). Exposes
derive_morphology_zone as pub for the integration test.

NOTE: golden file not yet generated + fixtures predate the T-1033 BodyParams strip;
fixed in follow-up commits after merging main.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Merge remote-tracking branch 'origin/main' into verification-harness b8f8a5fb23
test(simulation): align T-1031 harness with D-240 class-envelope temps de5f0f1e58 
Adapt the believability/budget verification harness to the T-1033 BodyParams
strip: validation-body fixtures (Kallast/Glødberg/Marevna) no longer set the
removed orbit/star fields (orbital_period_days, axial_tilt_deg, spectral_class,
star_type). Temperature is now derived from the planet_class envelope (D-240),
so the Marevna validation note/assertion is rewritten — oceanic band [-12,28]°C
+ maritime 0.6 + standard-atmosphere greenhouse yields ~+26°C at the equator
(was a stale ~-56°C orbit-equilibrium note). Add the seed→tile golden fixture
(tests/golden/derivation_harness.json) that the determinism regression locks.

Also fmt the harness and fix one clippy lint (iter().copied().collect() ->
to_vec()) — the file predated the pre-push gate's cargo-test/clippy coverage.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jpmschweitzer commented 2026-06-08 22:29:18 +02:00
Author
Owner
Copy Link
Copy Source

Review: verification-harness → main (type: code)

Two code reviewers (Hoshe — code quality/QA, Tyre — architecture). Both APPROVE; Hoshe raised three actionable findings, all addressed in 5cc4ff6c.

Hoshe (code quality / test correctness): APPROVE

The harness covers all four stated domains; the golden is a real regression surface (double-derivation equality + JSON value comparison). Three silent-pass / dead-code issues:

# File Issue Resolution
1 derivation_harness.rs assert_drainage_monotonicity + the 3 sweep tests Helper returns early when a chunk has no wet tiles; sweeps couldn't guarantee any chunk exercised the law → could pass with zero assertions Helper now returns whether it actually checked; each sweep asserts ≥1 chunk hit the wet path. Fixed.
2 derivation_harness.rs law_drainage_monotonicity_fjord_floor_at_sea_level Single-row Y=32 probe may miss the trough → sea-level law silently skipped Scans the full 64×64 chunk and asserts the deep-water trough exists. Fixed.
3 derivation_harness.rs make_test_terrain_analysis Dead code behind #[allow(dead_code)] + "future use" comment Removed. Fixed.

Tyre (architecture): APPROVE

The harness faithfully encodes the D-239 contract — the four §8 laws map onto real classifier behavior (not tautologies), the FjordWall GlaciationGrade ≥ 2 gate is tested on both sides against the actual gate, the golden pins the integer-discriminant surface per D-010, and the D-240 class-envelope alignment is sound. Making derive_morphology_zone pub is the correct minimal exposure (integration tests are a separate crate; pub(crate) would not link). All notes were informational/non-blocking:

  • Marevna temp comment (~+26°C) — verified against the runtime value (26.225°C); accurate. (Tyre's hand-calc of ~28°C assumed a flat greenhouse add + warm-end clamp that the implementation doesn't produce.)
  • +3 m levee tolerance and the fjord-floor ≤5 m asymmetry are correctly grounded in D-239 §9. No change.
  • Brittleness note: has_active_channel coupling may shift when Phase 4 BraidedPlain/lithology lands — harness degrades gracefully; re-confirm golden cases hit the wet path then. (Informational.)

Verdict: APPROVED (findings addressed in 5cc4ff6c)

Closes T-1031; completes epic T-974. Pre-push gate green (fmt, clippy -D warnings, full cargo test).

## Review: verification-harness → main (type: code) Two code reviewers (Hoshe — code quality/QA, Tyre — architecture). Both **APPROVE**; Hoshe raised three actionable findings, all addressed in `5cc4ff6c`. ### Hoshe (code quality / test correctness): APPROVE The harness covers all four stated domains; the golden is a real regression surface (double-derivation equality + JSON value comparison). Three silent-pass / dead-code issues: | # | File | Issue | Resolution | |---|------|-------|------------| | 1 | `derivation_harness.rs` `assert_drainage_monotonicity` + the 3 sweep tests | Helper returns early when a chunk has no wet tiles; sweeps couldn't guarantee any chunk exercised the law → could pass with zero assertions | Helper now returns whether it actually checked; each sweep asserts ≥1 chunk hit the wet path. **Fixed.** | | 2 | `derivation_harness.rs` `law_drainage_monotonicity_fjord_floor_at_sea_level` | Single-row `Y=32` probe may miss the trough → sea-level law silently skipped | Scans the full 64×64 chunk and asserts the deep-water trough exists. **Fixed.** | | 3 | `derivation_harness.rs` `make_test_terrain_analysis` | Dead code behind `#[allow(dead_code)]` + "future use" comment | Removed. **Fixed.** | ### Tyre (architecture): APPROVE The harness faithfully encodes the D-239 contract — the four §8 laws map onto real classifier behavior (not tautologies), the FjordWall `GlaciationGrade ≥ 2` gate is tested on both sides against the actual gate, the golden pins the integer-discriminant surface per D-010, and the D-240 class-envelope alignment is sound. Making `derive_morphology_zone` `pub` is the correct minimal exposure (integration tests are a separate crate; `pub(crate)` would not link). All notes were informational/non-blocking: - Marevna temp comment (~+26°C) — verified against the **runtime** value (`26.225°C`); accurate. (Tyre's hand-calc of ~28°C assumed a flat greenhouse add + warm-end clamp that the implementation doesn't produce.) - `+3 m` levee tolerance and the fjord-floor `≤5 m` asymmetry are correctly grounded in D-239 §9. No change. - Brittleness note: `has_active_channel` coupling may shift when Phase 4 BraidedPlain/lithology lands — harness degrades gracefully; re-confirm golden cases hit the wet path then. (Informational.) ### Verdict: APPROVED (findings addressed in `5cc4ff6c`) Closes **T-1031**; completes epic **T-974**. Pre-push gate green (fmt, clippy `-D warnings`, full `cargo test`).
jpmschweitzer added 1 commit 2026-06-08 22:36:14 +02:00
test(simulation): address PR #165 review (T-1031) 5cc4ff6c0c 
Hoshe review findings — close silent-pass gaps in the §8 drainage tests:
- assert_drainage_monotonicity now returns whether it actually checked (chunk
  had both wet and dry tiles); the three sweep tests (Alluvial/Meander/Braided
  Delta) assert at least one chunk exercised the law, so a regression that
  zeroes all channels fails loudly instead of passing vacuously.
- The fjord-floor test scans the full 64x64 chunk instead of a single Y=32 row
  and asserts the deep-water trough exists, so the sea-level floor law can no
  longer be skipped by a probe that missed the trough.
- Remove the unreferenced make_test_terrain_analysis helper (was behind
  #[allow(dead_code)] with a 'future use' comment — the maintenance trap the
  review flagged); make_dry_terrain_analysis covers the validation bodies.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
jpmschweitzer closed this pull request 2026-06-08 22:38:59 +02:00

Pull request closed

This pull request cannot be reopened because the branch was deleted.
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
jpmschweitzer (Jeroen Schweitzer)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: jpmschweitzer/settled-reach#165