feat(composer): the parameter roll -- infrastructure real, corpus starved

Tier 4a. The first piece of the composer that INVENTS rather than fixes.

The design, from COMPOSER-NOTES §3.3: do NOT roll 73 parameters
independently. The coherent region of a 73-dimensional space is a
vanishingly thin sheet and uniform sampling never lands on it -- and
independent rolling destroys the CORRELATIONS (room spacing <-> room
radius <-> tunnel length) that make a vector coherent. So blend known-good
vectors through the existing FStrateGenerationParams::Lerp instead:
constraint satisfaction is free, because a convex combination preserves
every relation holding in both parents. Min <= Max stays true because it
was never broken.

Built: FVoxelStrateCorpus (loads authored vectors, computes per-field
spread), VF_RollStrateParams (2-3 weighted parents, convex blend, jitter
by +-15% of each field's MEASURED corpus range -- not by the clamps, which
are "don't type nonsense" limits, not "this makes a good world" limits),
and VoxelForge.Composer.ParameterRoll. Determinism asserted by rolling
twice and comparing every field.

Exclusion list found by reflection, not assumption: StrateTopWorldZ and
StrateBottomWorldZ (manager-owned runtime bounds) plus 34 terrain-op
transport fields (TerraceStepHeight..PinchLength) that are copied in from
room operators rather than being strate tunables. Bools inherit from the
dominant parent; ints/enums use Lerp's existing snap.

⚠️ THE RESULT IS VACUOUS AND SAYS SO. 64/64 candidates "survived" -- but
the corpus loaded ONE vector (DA_Strate3), so all 64 are bit-identical.
100% survival of a single point is not evidence of anything. The box-
verdict check is vacuous too: 2,560 boxes, 0 proved, 0 voxels brute-forced,
so §6.2 is NOT closed. The test emits a warning saying exactly this rather
than reporting a clean pass, and no workaround was added to hide it.

Four strate assets exist (DA_Strate1..4); the loader found one. Fixing the
corpus is the next task -- with one parent there is nothing to blend, and
every number above is a placeholder.

Also owed: clamp metadata is editor-only, so the safety net needs a
cook-time baked table before this ships.

Nothing calls the composer from any runtime path (§3.1: offline, once per
season, on a build box).

Build clean, 23 tests / 0 failed (1 carries the deliberate vacuity
warning), verified in VoxelM.log.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VJiTcTSDNkKBtnNdiBtdaU

CODEMAP.md

@@ -438,6 +438,17 @@ One asset = one biome: identity + `DebugColor`, climate placement box (`ReliefMi
 atmosphere override, `WaterMaterial`, `MaterialPaletteIndex` (F6 — baked to vertex colour, §8.15)), `GameplayTags`. Referenced from
 `UVoxelStrateDefinition::Biomes[]`. Generator-agnostic (surface biomes now, cave biomes later). §8.14.
 
+**`Public/VoxelStrateComposer.h` + `Private/VoxelStrateComposer.cpp`** (NEW, Tier 4a) — offline
+`FVoxelStrateCorpus` and `VF_RollStrateParams`. Walks the settings asset's fixed/pool definition
+references, admits only `TunnelNetwork` / `Underwater` vectors (the archetypes that author
+`GenerationParams`), measures every X-macro field's min/max/mean/population stddev, and exposes the
+explicit non-tunable list. Rolls are pure in `(corpus contents, seed, index)`: weighted 2–3-parent
+same-archetype selection → `FStrateGenerationParams::Lerp` → ±15% of measured field range jitter →
+editor-reflection clamps → ordered-pair repair. Bool `bTunnelsFlowTowardOrigin` inherits from the
+dominant parent; integer/enum SNAP fields are not jittered. This API is not called by runtime
+generation. The first project corpus run found one eligible `DA_Strate3` vector, so all measured
+spreads were zero; that limitation is recorded by the Tier 4a test/report rather than hidden.
+
 ### 3.9 Player edits — `Public/VoxelDiffLayer.h` + `.cpp`
 `UVoxelDiffLayer : UObject` (h:77). Stores `FVoxelModification` (h:43: Center/Radius/Strength;
 **negative Strength = carve, positive = fill**) grouped by chunk in `TMap ChunkMods`.
@@ -501,6 +512,7 @@ The plugin's first tests (`OPSTACK-PLAN.md` Phase 0.5). Run them from the editor
 | `VoxelForgeOpStackIslandTest.cpp` | `VoxelForge.OpStack.FloatingIslandEquivalence` | The port that runs the stack **backwards** — void + fill vs rock + carve, same classes with the opposite sign. Counts interior-solid and open-void samples separately (on this archetype an aggregate "N solid" is dominated by the seal bands and says nothing about the islands). Counts `AllSolid` and `AllAir` verdicts **separately** too: `AllAir` is the one no cave archetype could ever prove, and it is the entire perf argument here. |
 | `VoxelForgeOpStackMazeTest.cpp` | `VoxelForge.OpStack.MazeEquivalence` | **Phase 1's load-bearing test.** The 7-op Maze stack vs `GetMazeDensity` over 20k points (aiming for bit-identity; a side-of-iso disagreement is the hard fail), plus purity across workers and brute force on every box verdict the stack emits. Reports how many tiles the stack can prove uniform — today's `ClassifyTile` proves **zero** for any cave archetype. |
 | `VoxelForgeStrateParamCoverageTest.cpp` | `VoxelForge.Determinism.StrateParamBlendCoverage` | **The X-macro guard** (added 2026-08-17). `FStrateGenerationParams::Lerp` blends the hand-written `VF_STRATE_PARAM_FIELDS` list, **not** the struct — so a field added to one and not the other compiles, tests green, and silently takes its **default** inside every Gradient/Interleaved transition band. This expands the X-macro a **third** way (after LERP and SNAP), into a name list, and diffs it against the struct's UObject reflection. Pure shape test: no fixture, no world, instant. `GExemptFieldNames` is **empty** — every reflected field is covered today, and any exemption must be written down as a decision. Stakes rise with the world composer, which intends to invent parameter sets through this same `Lerp` (`COMPOSER-NOTES.md`). |
+| `VoxelForgeComposerParameterRollTest.cpp` | `VoxelForge.Composer.ParameterRoll` | Loads the hand-authored settings corpus, prints the complete per-field spread/exclusion/clamp table, asserts deterministic bit-identical rolls, measures 64 transient candidate strates with `VF_MeasureStrate` plus the exact unsnapped arrival→departure law, and brute-forces every rolled production box verdict. The first run found one eligible vector (all spreads zero), 64/64 survival, and a deliberately reported vacuous §6.2 scan: 2,560 Mixed, 0 proved, 0 violations. |
 | `VoxelForgeLayoutOrderIndependenceTest.cpp` | `VoxelForge.Determinism.LayoutOrderIndependence` | Builds a known transient soft-pointer pool, then rebuilds it in original, reversed, and swapped orders. Requires a non-empty layout and passage set, and compares every slot's definition/Z/height plus passage endpoints, radius, type, control geometry, and bounds bit-for-bit. |
 | `VoxelForgePassageOpenSpaceTest.cpp` | `VoxelForge.Determinism.PassageLandsInOpenSpace` | Uses the real fixture density path to check every generated inter-strate passage whose destination query answers: a 16-point ring outside the mouth's carve/blend band has at least half its samples in destination air, and the endpoint matches the pure open-point result within the mouth's float envelope. This is a connectivity proxy, not a flood-fill proof. Reports checked passages and false/unanswerable archetypes; fails if it inspects zero passages. |
 | `VoxelForgeStrateConnectivityTest.cpp` | `VoxelForge.Generation.StrateConnectivity` / `VoxelForge.Generation.StrateConnectivityRefinement` / `VoxelForge.Generation.VerticalShaftSeamFreedom` | Bounded strate metrics with density-polarity and solid-gap controls, deterministic route rechecks, and refinement sweeps. The refinement test fits the measurement AABB to each arrival/departure mouth pair, reports exact before/after seed-6 cell counts, reruns every negative with doubled margin, measures tree orphan candidates/path reachability and roughness-bubble proxies, asserts axis landings, checks source/mirror physical paths, and requires 16/16 effective arrival→departure results across the 16 VerticalShafts seeds. Margin-binding negatives are reported as measurement limits, never as gap findings. The seam test re-evaluates cell-boundary positions after warming distinct neighbouring chunk contexts in both legacy and operator-stack paths. |