settled-reach/docs/workshops/tile-derivation-contract/workshop-outcomes.md

title, description, type, status, workshop, created, decision_refs, resolves, participants

title	description	type	status	workshop	created	decision_refs	resolves	participants
Workshop Outcomes: Tile Derivation Contract (Q-101)	Final outcomes of the tile-derivation-contract workshop — resolution of Q-101 via D-239, the coarse→fine tile refinement chain, morphology family, anti-squaring warp, and district-temperature climate primitive.	workshop	complete	tile-derivation-contract	2026-06-07	D-239	Q-101	Tyre Gestalt Troblum Miri Qatux

Workshop Outcomes: Tile Derivation Contract (Q-101)

Result: Q-101 resolved → D-239 filed (architecture domain, confirmed, 2026-06-07).

The keystone "what's in a tile" question — how a coarse Layer-1 cell + seed becomes coherent ~1 m voxel geometry across ~3 scale jumps — is now contracted. This was the one unbuilt layer of the Phase-4 cascade (L0–L4 already exist).

What was decided (D-239 in brief)

1. Three-carrier refinement chain. RegionProfile (~1 km) → ChunkContext (64 m) → VoxelColumn (1 m). Each a pure deterministic function of (seed, atlas, body-params, position) per D-227. Each scale boundary is its own derivation.

2. No authoring at the derivation layers, ever. Zero per-body override hooks. All gating params (RIVER_THRESHOLD → derived per-body-class, not the global 200; tectonic_class; GlaciationGrade; precipitation_class) derive from body params. Lore bodies (Kallast, Velen, Cygni B) are honoured by setting params and serve as validation cases — fix params, never patch the derivation.

3. District temperature is the climate primitive. A scalar °C per 2×2 km district, nullable (no atmosphere → null → no climate branch; airless surface ice is geology). Derived from sun (luminosity + insolation), planet (latitude + elevation lapse + orbital/axial phase = season, day-phase = diurnal), atmosphere (greenhouse → base, heat-retention → diurnal-swing amplitude). Moisture is a separate primitive. Everything climatic — precipitation, vegetation/treeline, glaciation, the seasonal/clock state (Q-105) — derives from temperature (+ moisture). Formalises D-210's temperature proxy.

4. Scattered, transient freeze/snow model. Ice/snow is a band with coherent spatial scatter, not a hard temp<0 contour: freshwater (lakes/rivers) +5 → −10 °C district mean (the +5 is night-frost reaching the coldest tiles); sea ice its own lower/wider band (≈ −2 onset, pack-ice pattern); snow (land) the same, gated on moisture. Transient in the marginal band — forms in the cold phase (night/season), melts in the warm phase → time-of-day passability shifts. Forward contract to Q-105.

5. Anti-"squaring" warp. A stateless, hash-based position-keyed domain-warp; f64 sub-metre precision through to the final voxel coordinate, then truncate to the integer voxel address (a cast, IEEE-754-deterministic). ±8 m range makes FMA ULP variance harmless. Position math, not a structural decision — D-010 integer discipline preserved.

6. 8 morphology families via a gated decision tree over integer RegionProfile inputs: LavaField · FjordWall · CliffCoast · BraidedDelta · DuneStrand · IncisedGorge · MeanderReach · AlluvialPlain (fallback). Hard gates: fjord ≥ GlaciationGrade 2; LavaField = Volcanic; lithology bounds slope/form.

7. Frozen 17-zone MorphologyZone vocabulary. Freeze point = the Rust enum; changes need a D-record amendment (classifier tuning stays free). Four zones (tidal flat, estuarine, alpine, wetland) are derived sub-classifications, not generator families.

8. Seams — prevent incompatible, allow valid (never patch). Decision-tree gate ordering

   • a build-time compatibility-matrix invariant blocks incompatible adjacencies (e.g. MeanderReach↔Volcanic). Valid geomorphic seams (cliff↔fjord, lithology faults) are kept sharp (real geology is sharp), made organic by the warp — no feathering.

9. Believability laws + game-feel constraints are binding (drainage monotonicity, lithology→landform, glaciation→form, climate→vegetation; ≥1 tactical point/chunk, cover ≠ concealment, chokepoint widths, real seasonal passability via ElevationDelta↔Q-105).

Process notes

• 2-round lean format (independent positions → lead-interviewed synthesis), not the full 5-round workshop. Round-1 positions: tyre-round1.md, gestalt-round1.md, troblum-round1.md, miri-round1.md.
• The four core calls (seams, body-params-fully-derived, vocabulary lock, warp precision) were made by the human via AskUserQuestion. The climate model (temperature primitive, separate moisture, the three-sibling scattered freeze, day/night swing, transient ice/snow) was a human-driven design pivot during synthesis.
• An adversarial verification pass (the same four perspectives) caught ~10 issues, all folded into the final D-239: the "continuous→continuous" seam argument was false (boolean gates are discontinuities) → rewritten to gate-ordering + matrix; warp determinism gaps (FMA / statelessness / truncate-not-compare); "Block-or-higher" → "Region-or-higher"; D-209 tier → D-203; missing LavaField tectonic gate; seasonal regained mechanical teeth; the 17-zones-vs-8-families gap closed; believability laws inlined; "frozen" given a real enforcement mechanism.

Open siblings (left for later, contract does not preclude them)

• Q-102 — cohesion matrix: effectively is the D-239 warp; can be closed against it or folded in when the material-scatter layer is built.
• Q-103 — tile-mutator op schema: open, needed at the save system (Phase 5+).
• Q-105 — region seasonal/clock state: D-239's temperature(time) + ElevationDelta calibration is the forward contract to it.

Process friction captured

• Team mode is broken in clide — see docs/briefings/clide-team-mode-friction.md. Teammates ran in-process (no pane), the Team Chat view stayed inert, no clide team verb. The workshop ran as background Agent subagents instead. Hand-off material for the clide team.
• The miri agent type is read-only (no Write/Bash) — the lead transcribed miri-round1.md to disk.