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

fix(simulation): address believability-stack review — all findings processed

Browse Source 
Stack review (Hoshe + Tyre) of T-1083+T-1082+T-1080. Every finding fixed or disposed:

- believability.rs sample_indices: replace the rejection-sampling loop (could stall
  when n ≈ take) with deterministic partial Fisher-Yates — O(n), guaranteed
  termination, distinct, sorted (Hoshe+Tyre).
- believability.rs drainage proxy: exclude water-body districts (all elev 0, no banks)
  and require both wet + dry voxels — they were passing vacuously and inflating the
  metric (Hoshe).
- believability.rs read_cities: log skipped malformed rows instead of dropping silently
  (Hoshe).
- voxel.rs: add the 3 WaterBody zones to all_families_no_panic_at_extreme_positions +
  fix the "7 families" comment (Hoshe).
- climate_constants.toml: bold warning that [moisture_gradient] is not loaded until
  T-1032 — tune ClimateConstants::default() (Tyre).
- Q-123: record the per-zone "water renders wet" threshold + the land-relative
  vegetation-present denominator as calibration items (Tyre + the open 7th criterion).
- T-1083 ticket: note the sampling-fix attribution (landed in the T-1082 commit).

Believability golden regenerated (sampler + drainage fixes moved the metrics).
clippy --all-targets -D warnings clean; 1580 lib tests + harness pass.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
jpmschweitzer
2026-06-28 17:41:03 +02:00
co-authored by Claude Opus 4.8
parent bb5069aa13
commit f771c9ce4e
7 changed files with 99 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.pql/changelog/ticket_history/2026-06.sql
+29
View File
@@ -2739,3 +2739,32 @@ TUNING: add [moisture_gradient] to server/data/climate_constants.toml + matching
DETERMINISM: integer (water_dist u16, integer division, clamp 0..100). D-010 ok.
VERIFY: believability harness gate moisture_q.distinct>=3 flips FAIL→PASS; vegetation-present should rise too. Magnitude calibration is Q-123 (provisional coefficients now). Files: district_profile.rs, climate_constants.toml.', NULL, '2026-06-28 13:50:47', '2026-06-28 13:50:47.912', '2026-06-28 13:50:47.912', NULL, '47bb1324b99af91b19f50ee8f8d3825a', 2) ON CONFLICT(hash) DO NOTHING;
INSERT INTO ticket_history (ticket_record_id, field, old_value, new_value, changed_by, changed_at, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('06FGSJZG3WFPQ3VXBFQFM4YEWW', 'description', 'Promote the throwaway probe (server/src/bin/aliveness_probe.rs) into a committed, repeatable believability test protocol — the instrument that found T-1080/T-1081/T-1082. For a body+seed it runs the deterministic cascade and emits: (1) cross-district CONTRAST/variance metrics — moisture, elevation/relief, slope, morphology-zone diversity, terrain-material diversity, vegetation gradient — the screen that catches ''uniform/dead'' worlds; (2) COHERENCE checks with pass/fail — water zones must render wet, vegetation must correlate with moisture/water-distance/slope/aspect, drainage monotonic; (3) optional per-layer PNG maps (a stopgap for the T-960 Atlas viewer). LESSON BAKED IN: the first naive ''is it vegetated'' metric reported a broken uniform world as ALIVE — so the protocol MUST measure CONTRAST + COHERENCE across samples, never marginal per-tile counts. Make it runnable per-body, deterministic (string seed -> u64), snapshot a baseline for regression. This is the D-245 ENFORCER. NOTE: today it samples at the DISTRICT tier because no production code maps a world chunk -> covering DistrictProfile yet (voxel layer is a walking skeleton). Distinct from T-1031 (binding-LAW harness); believability sibling. Relates to T-960.
---
Implemented 2026-06-28 (branch believability-harness).
- server/src/atlas/believability.rs (new): the enforcer core. BelievabilityReport (serde, golden-able) + analyze() = CONTRAST (per-field min/max/distinct over ALL districts for moisture/elev/slope/ocean; distinct morphology-zones/vegetation-classes/terrain-materials) + COHERENCE (water-renders-wet [T-1082], drainage-monotonic, vegetation-present). evaluate_criteria() = advisory D-245 checks (thresholds Q-123-TBD). cascade_for_body()/seed_to_u64() loader shared by bin + harness. Unit tests prove the metric distinguishes uniform (fails) vs varied (passes) — the ''never marginal per-tile counts'' lesson — + determinism.
- server/src/bin/aliveness_probe.rs: refactored to a thin CLI over the module; prints the body-level report + advisory D-245 criteria; dropped the naive per-tile ''reads ALIVE'' verdict.
- server/tests/believability_harness.rs (+ tests/golden/believability.json): real cascade for Arbour (temperate/ocean) + Edict (frozen/ice); determinism assert; golden regression (updates when T-1080/T-1082 land); BELIEVABILITY_STRICT=1 fails on unmet criteria. Skips if data absent. x86_64 golden.
BASELINE @ believability-v1: Arbour 3/7, Edict 3/7. FAIL: moisture gradient (distinct=1, T-1080), water-renders-wet (T-1082), drainage-monotonic. Edict vegetation 0/64 = blank tundra (the frozen ''reads dead'' case). PASS: elevation relief, morphology variety, vegetation-present (Arbour only).
DEFERRED (ticket item 3 — explicitly optional): per-layer PNG maps (the T-960 Atlas-viewer stopgap). The contrast+coherence metrics are the core enforcer; maps are a follow-up if the human-gate review wants them before T-960.
This is the D-245 enforcer; advisory now, ratchets to strict (BELIEVABILITY_STRICT) once Q-123 calibrates thresholds and T-1080/1081/1082 flip the criteria.', 'Promote the throwaway probe (server/src/bin/aliveness_probe.rs) into a committed, repeatable believability test protocol — the instrument that found T-1080/T-1081/T-1082. For a body+seed it runs the deterministic cascade and emits: (1) cross-district CONTRAST/variance metrics — moisture, elevation/relief, slope, morphology-zone diversity, terrain-material diversity, vegetation gradient — the screen that catches ''uniform/dead'' worlds; (2) COHERENCE checks with pass/fail — water zones must render wet, vegetation must correlate with moisture/water-distance/slope/aspect, drainage monotonic; (3) optional per-layer PNG maps (a stopgap for the T-960 Atlas viewer). LESSON BAKED IN: the first naive ''is it vegetated'' metric reported a broken uniform world as ALIVE — so the protocol MUST measure CONTRAST + COHERENCE across samples, never marginal per-tile counts. Make it runnable per-body, deterministic (string seed -> u64), snapshot a baseline for regression. This is the D-245 ENFORCER. NOTE: today it samples at the DISTRICT tier because no production code maps a world chunk -> covering DistrictProfile yet (voxel layer is a walking skeleton). Distinct from T-1031 (binding-LAW harness); believability sibling. Relates to T-960.
---
Implemented 2026-06-28 (branch believability-harness).
- server/src/atlas/believability.rs (new): the enforcer core. BelievabilityReport (serde, golden-able) + analyze() = CONTRAST (per-field min/max/distinct over ALL districts for moisture/elev/slope/ocean; distinct morphology-zones/vegetation-classes/terrain-materials) + COHERENCE (water-renders-wet [T-1082], drainage-monotonic, vegetation-present). evaluate_criteria() = advisory D-245 checks (thresholds Q-123-TBD). cascade_for_body()/seed_to_u64() loader shared by bin + harness. Unit tests prove the metric distinguishes uniform (fails) vs varied (passes) — the ''never marginal per-tile counts'' lesson — + determinism.
- server/src/bin/aliveness_probe.rs: refactored to a thin CLI over the module; prints the body-level report + advisory D-245 criteria; dropped the naive per-tile ''reads ALIVE'' verdict.
- server/tests/believability_harness.rs (+ tests/golden/believability.json): real cascade for Arbour (temperate/ocean) + Edict (frozen/ice); determinism assert; golden regression (updates when T-1080/T-1082 land); BELIEVABILITY_STRICT=1 fails on unmet criteria. Skips if data absent. x86_64 golden.
BASELINE @ believability-v1: Arbour 3/7, Edict 3/7. FAIL: moisture gradient (distinct=1, T-1080), water-renders-wet (T-1082), drainage-monotonic. Edict vegetation 0/64 = blank tundra (the frozen ''reads dead'' case). PASS: elevation relief, morphology variety, vegetation-present (Arbour only).
DEFERRED (ticket item 3 — explicitly optional): per-layer PNG maps (the T-960 Atlas-viewer stopgap). The contrast+coherence metrics are the core enforcer; maps are a follow-up if the human-gate review wants them before T-960.
This is the D-245 enforcer; advisory now, ratchets to strict (BELIEVABILITY_STRICT) once Q-123 calibrates thresholds and T-1080/1081/1082 flip the criteria.
Stack-review note (2026-06-28): the believability sampling-bias fix (contiguous-prefix -> seeded Fisher-Yates spread in sample_indices) is logically a T-1083 enforcer-correctness fix but landed in the T-1082 commit cdc55032d (found while verifying T-1082). git history records it accurately under the water-body work; flagged so a bisect lands in the expected place. The Fisher-Yates termination fix + drainage-proxy exclusion of water districts (stack review) are in the T-1080 review-fix commit.', NULL, '2026-06-28 15:38:18', '2026-06-28 15:38:18.309', '2026-06-28 15:38:18.309', NULL, '4b9f44267a786f912a1d4744ee1aa36b', 2) ON CONFLICT(hash) DO NOTHING;
.pql/changelog/tickets/2026-06.sql
+16
View File
@@ -3672,3 +3672,19 @@ TUNING: add [moisture_gradient] to server/data/climate_constants.toml + matching
DETERMINISM: integer (water_dist u16, integer division, clamp 0..100). D-010 ok.
VERIFY: believability harness gate moisture_q.distinct>=3 flips FAIL→PASS; vegetation-present should rise too. Magnitude calibration is Q-123 (provisional coefficients now). Files: district_profile.rs, climate_constants.toml.', 'in_progress', 'high', NULL, 'server', 'D-239', '2026-06-28 05:54:42.258', '2026-06-28 13:50:47.912', NULL, '4f5ee94f5fd8f5e447968619a520f915', 2) ON CONFLICT(record_id) DO UPDATE SET type=excluded.type, parent_record_id=excluded.parent_record_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);
INSERT INTO tickets (record_id, type, parent_record_id, title, description, status, priority, assigned_to, team, decision_ref, created_at, updated_at, deleted_at, hash, canonical_version) VALUES ('06FGSJZG3WFPQ3VXBFQFM4YEWW', 'story', '06FGSJNP8YM537DJCC82ZD6544', 'Repeatable believability test protocol — cross-district contrast + coherence metrics (promote aliveness_probe)', 'Promote the throwaway probe (server/src/bin/aliveness_probe.rs) into a committed, repeatable believability test protocol — the instrument that found T-1080/T-1081/T-1082. For a body+seed it runs the deterministic cascade and emits: (1) cross-district CONTRAST/variance metrics — moisture, elevation/relief, slope, morphology-zone diversity, terrain-material diversity, vegetation gradient — the screen that catches ''uniform/dead'' worlds; (2) COHERENCE checks with pass/fail — water zones must render wet, vegetation must correlate with moisture/water-distance/slope/aspect, drainage monotonic; (3) optional per-layer PNG maps (a stopgap for the T-960 Atlas viewer). LESSON BAKED IN: the first naive ''is it vegetated'' metric reported a broken uniform world as ALIVE — so the protocol MUST measure CONTRAST + COHERENCE across samples, never marginal per-tile counts. Make it runnable per-body, deterministic (string seed -> u64), snapshot a baseline for regression. This is the D-245 ENFORCER. NOTE: today it samples at the DISTRICT tier because no production code maps a world chunk -> covering DistrictProfile yet (voxel layer is a walking skeleton). Distinct from T-1031 (binding-LAW harness); believability sibling. Relates to T-960.
---
Implemented 2026-06-28 (branch believability-harness).
- server/src/atlas/believability.rs (new): the enforcer core. BelievabilityReport (serde, golden-able) + analyze() = CONTRAST (per-field min/max/distinct over ALL districts for moisture/elev/slope/ocean; distinct morphology-zones/vegetation-classes/terrain-materials) + COHERENCE (water-renders-wet [T-1082], drainage-monotonic, vegetation-present). evaluate_criteria() = advisory D-245 checks (thresholds Q-123-TBD). cascade_for_body()/seed_to_u64() loader shared by bin + harness. Unit tests prove the metric distinguishes uniform (fails) vs varied (passes) — the ''never marginal per-tile counts'' lesson — + determinism.
- server/src/bin/aliveness_probe.rs: refactored to a thin CLI over the module; prints the body-level report + advisory D-245 criteria; dropped the naive per-tile ''reads ALIVE'' verdict.
- server/tests/believability_harness.rs (+ tests/golden/believability.json): real cascade for Arbour (temperate/ocean) + Edict (frozen/ice); determinism assert; golden regression (updates when T-1080/T-1082 land); BELIEVABILITY_STRICT=1 fails on unmet criteria. Skips if data absent. x86_64 golden.
BASELINE @ believability-v1: Arbour 3/7, Edict 3/7. FAIL: moisture gradient (distinct=1, T-1080), water-renders-wet (T-1082), drainage-monotonic. Edict vegetation 0/64 = blank tundra (the frozen ''reads dead'' case). PASS: elevation relief, morphology variety, vegetation-present (Arbour only).
DEFERRED (ticket item 3 — explicitly optional): per-layer PNG maps (the T-960 Atlas-viewer stopgap). The contrast+coherence metrics are the core enforcer; maps are a follow-up if the human-gate review wants them before T-960.
This is the D-245 enforcer; advisory now, ratchets to strict (BELIEVABILITY_STRICT) once Q-123 calibrates thresholds and T-1080/1081/1082 flip the criteria.
Stack-review note (2026-06-28): the believability sampling-bias fix (contiguous-prefix -> seeded Fisher-Yates spread in sample_indices) is logically a T-1083 enforcer-correctness fix but landed in the T-1082 commit cdc55032d (found while verifying T-1082). git history records it accurately under the water-body work; flagged so a bisect lands in the expected place. The Fisher-Yates termination fix + drainage-proxy exclusion of water districts (stack review) are in the T-1080 review-fix commit.', 'in_progress', 'high', NULL, 'server', 'D-239', '2026-06-28 05:55:19.712', '2026-06-28 15:38:18.308', NULL, '0b6572cca4a24c9658cf4751d7573add', 2) ON CONFLICT(record_id) DO UPDATE SET type=excluded.type, parent_record_id=excluded.parent_record_id, title=excluded.title, description=excluded.description, status=excluded.status, priority=excluded.priority, assigned_to=excluded.assigned_to, team=excluded.team, decision_ref=excluded.decision_ref, updated_at=excluded.updated_at, deleted_at=excluded.deleted_at, hash=excluded.hash, canonical_version=excluded.canonical_version WHERE excluded.updated_at > tickets.updated_at OR (excluded.updated_at = tickets.updated_at AND excluded.hash > tickets.hash);