feat(composer): it invents strates that work -- 59/64, and §6.2 is CLOSED

The corpus was starved: the loader followed the active settings pool and found ONE vector, so all 64 candidates were bit-identical and every number was a placeholder. It now loads every UVoxelStrateDefinition via the Asset Registry (Saved/Autosaves and Saved/Cooked excluded -- including them would silently weight the corpus toward whatever was autosaved last) plus the per-archetype defaults from VoxelStrateTypes.h, which are themselves hand-tuned known-good vectors. Blending stays WITHIN an archetype; different archetypes read different subsets of the 73 fields. ⭐ THE COMPOSER WORKS. 64 candidates blended from real parents, every one measured: survival 59/64 (92.2%) non-vacuous 60 · largest-share >= .50 60 · primordial law 63 · roll failures 0 and they are genuinely different worlds, not one world 64 times: air fraction .0137 -> 1.000 walkable .000 -> .389 feature scale 4 -> 376 voxels archetypes TunnelNetwork, VerticalShafts, FlatPlain, CrystalChamber, SurfaceWorld, Underwater, FloatingIslands, Maze ⭐⭐ §6.2 IS CLOSED, and it was the real risk. Box-verdict bounds were only ever proved against the parameters Jahni typed; the whole worry was that invented parameters would violate a bound nobody had checked -- silently, leaving no geometry and no collision. Against ROLLED parameters: 1,224 boxes proved (549 AllSolid, 675 AllAir), 1,336 Mixed 1,629,144 voxels brute-forced 0 VIOLATIONS The bounds hold across the rolled range, not just at authored points. The isolation work in 8e346a3 is what made that true. The 5 rejections are a real finding, not noise: FloatingIslands can jitter its islands out of existence and return an all-air strate (air 1.000, walkable 0.000, feature scale 0). That is exactly what reject-and-resample (§3.5) is for -- the measurement caught it without anyone anticipating it, which is the whole argument for measuring rather than trusting. Roll+measure 112.6 s for 64 candidates. §3.1's budget is "runs overnight", so there is room for thousands. Build clean, 23 succeeded / 0 warnings / 0 failed (the deliberate vacuity warning is gone because the check is no longer vacuous). Verified by me. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VJiTcTSDNkKBtnNdiBtdaU
2026-09-04 08:32:01 +02:00
1 parent 358bd415a9
commit 190ff381c0
@@ -438,16 +438,21 @@ 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.
+**`Public/VoxelStrateComposer.h` + `Private/VoxelStrateComposer.cpp`** (Tier 4a) — offline
+`FVoxelStrateCorpus` and `VF_RollStrateParams`. `LoadFromAssetRegistry` enumerates every project
+`UVoxelStrateDefinition` through the Asset Registry, excludes `Saved/Autosaves` and `Saved/Cooked`
+copies, and adds one `VoxelStrateTypes.h` default vector for each of the eight exact archetypes.
+The corpus is grouped by exact `ECaveGeneratorType`: the 2026-09-04 run loaded **4 project
+vectors + 8 defaults = 12 members** in **8 groups**. Sibling families use their native structs
+(`FSlabGenerationParams`, `FMazeGenerationParams`, `FSurfaceGenerationParams`,
+`FVerticalShaftParams`, `FFloatingIslandParams`); no cross-archetype blend is attempted.
+Spreads are measured per group, including the plain-native tunnel transport fields for reporting.
+Rolls are pure in `(corpus contents, seed, index)`: weighted 2–3-parent same-archetype selection →
+`FStrateGenerationParams::Lerp` or reflected native-family blend → ±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 settings audit found **one unique path**, `/Game/VoxelForge/DA_Strate3`,
+which explains the old one-vector corpus; the project assets were never in that pool.

 ### 3.9 Player edits — `Public/VoxelDiffLayer.h` + `.cpp`
 `UVoxelDiffLayer : UObject` (h:77). Stores `FVoxelModification` (h:43: Center/Radius/Strength;
@@ -512,7 +517,7 @@ 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. |
+| `VoxelForgeComposerParameterRollTest.cpp` | `VoxelForge.Composer.ParameterRoll` | Asset-Registry corpus audit + complete per-archetype spread/exclusion/clamp table; asserts bit-identical deterministic rerolls and same-archetype parents; measures 64 transient candidate strates with `VF_MeasureStrate` plus the exact unsnapped arrival→departure law; brute-forces every rolled production box verdict. The final full-suite run: **59/64 survival (92.2%)**, **1,336 Mixed + 549 AllSolid + 675 AllAir = 1,224 proved boxes**, **1,629,144 lattice voxels checked, 0 violations**, 112.574 s total. |
 | `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. |
@@ -200,18 +200,66 @@ exclusion list. Bools cannot be blended — roll them by probability or inherit

 `Public/VoxelStrateComposer.h` + `Private/VoxelStrateComposer.cpp` now implement the corpus and
 the deterministic roll. `Private/Tests/VoxelForgeComposerParameterRollTest.cpp` is the offline
-measurement harness. The roll uses `FStrateGenerationParams::Lerp`, chooses 2–3 weighted parents
-inside one archetype group, applies jitter of **±15% of each field's measured corpus range
-(`max-min`)**, then applies reflected `ClampMin`/`ClampMax` metadata. It does not use clamps as
-distribution ranges, does not touch generation, and leaves `WorldRadiusVoxels` at its default 0.
+measurement harness. The roll uses the native family blend (`FStrateGenerationParams::Lerp` for
+the tunnel family), chooses 2–3 weighted parents **inside one exact archetype group**, applies
+jitter of **±15% of each field's measured corpus range (`max-min`)**, then applies reflected
+`ClampMin`/`ClampMax` metadata. It does not use clamps as distribution ranges, does not touch
+generation, and leaves `WorldRadiusVoxels` at its default 0. The sibling native parameter structs
+are carried in the same corpus entry, so a `SurfaceWorld` vector is never blended with a
+`TunnelNetwork` vector merely because both happen to be stored in a strate asset.

-The loader walks the settings asset's fixed-strate and pool references, de-duplicates by asset path,
-and accepts only `TunnelNetwork` / `Underwater`: those are the only archetypes whose authored vector
-is `GenerationParams`. The current `DA_Settings` asset contains **one unique resolved reference**,
-`DA_Strate3`, and it is `TunnelNetwork`; the other authored archetype assets are not in that
-settings corpus. Consequently every measured field has zero spread, all 64 rolls repeat the same
-known-good vector (parent weights still roll deterministically), and this run did not exercise
-non-zero jitter. That is a corpus-size finding, not a rate optimization.
+The original one-vector result was caused by the old loader following `DA_Settings`' fixed-strate
+and pool references. An audit of those references finds **one unique path**,
+`/Game/VoxelForge/DA_Strate3.DA_Strate3`, whose archetype is `TunnelNetwork`; that is why the
+previous corpus only loaded one asset. The current loader instead asks the Asset Registry for every
+project `UVoxelStrateDefinition`, sorts and de-duplicates the results, accepts `/Game` packages,
+and ignores `Saved/Autosaves` and `Saved/Cooked` copies. The scan found **4 project assets**, with
+no saved-copy, outside-project, duplicate, or unresolved records:
+
+| Asset | Archetype |
+|---|---|
+| `DA_Strate1` | `SurfaceWorld` |
+| `DA_Strate2` | `FloatingIslands` |
+| `DA_Strate3` | `TunnelNetwork` |
+| `DA_Strate4` | `TunnelNetwork` |
+
+The hand-authored C++ defaults in `VoxelStrateTypes.h` are also corpus members: one for each of the
+8 supported archetypes (`TunnelNetwork`, `FlatPlain`, `CrystalChamber`, `Maze`, `SurfaceWorld`,
+`VerticalShafts`, `FloatingIslands`, `Underwater`). The fed corpus is therefore **12 members** in
+**8 exact-archetype groups**. Group membership is `TunnelNetwork=3`, `FlatPlain=1`,
+`CrystalChamber=1`, `Maze=1`, `SurfaceWorld=2`, `VerticalShafts=1`, `FloatingIslands=2`, and
+`Underwater=1`.
+
+The complete 253-descriptor spread table is emitted by the automation test, including sample count,
+min/max/mean/standard deviation, excluded/reflected status, and clamp metadata. The 230 zero-range
+rows are intentionally omitted below; these are the **23 rows with real spread** (float formatting
+is the test's measured output):
+
+| Archetype | Field | N | Min | Max | Mean | StdDev |
+|---|---|---:|---:|---:|---:|---:|
+| TunnelNetwork | MinRoomRadius | 3 | 10 | 25 | 15 | 7.07106781 |
+| TunnelNetwork | MaxRoomRadius | 3 | 30 | 125 | 61.6666667 | 44.7834295 |
+| TunnelNetwork | OriginRoomRadius | 3 | 20 | 120 | 53.3333333 | 47.1404521 |
+| TunnelNetwork | TunnelMinRadius | 3 | 3 | 9 | 5 | 2.82842712 |
+| TunnelNetwork | TunnelMaxRadius | 3 | 7 | 15 | 9.66666667 | 3.77123617 |
+| SurfaceWorld | BaseGroundRelative | 2 | 0.200000003 | 0.300000012 | 0.250000007 | 0.0500000045 |
+| SurfaceWorld | ElevationRange | 2 | 60 | 200 | 130 | 70 |
+| SurfaceWorld | ContinentFrequency | 2 | 0.00100000005 | 0.00600000005 | 0.00350000005 | 0.0025 |
+| SurfaceWorld | MountainStrength | 2 | 0.5 | 1 | 0.75 | 0.25 |
+| SurfaceWorld | MountainFrequency | 2 | 0.00200000009 | 0.0120000001 | 0.0070000001 | 0.005 |
+| SurfaceWorld | SurfaceRoughness | 2 | 2 | 3 | 2.5 | 0.5 |
+| SurfaceWorld | ReliefStrength | 2 | 0 | 0.699999988 | 0.349999994 | 0.349999994 |
+| SurfaceWorld | ReliefContrast | 2 | 1.60000002 | 2 | 1.80000001 | 0.199999988 |
+| SurfaceWorld | TerraceHeight | 2 | 12 | 20 | 16 | 4 |
+| SurfaceWorld | BeachWidth | 2 | 8 | 200 | 104 | 96 |
+| FloatingIslands | IslandSpacing | 2 | 95 | 1000 | 547.5 | 452.5 |
+| FloatingIslands | IslandDensity | 2 | 0.100000001 | 0.5 | 0.300000001 | 0.199999999 |
+| FloatingIslands | IslandMinRadius | 2 | 18 | 50 | 34 | 16 |
+| FloatingIslands | IslandMaxRadius | 2 | 42 | 500 | 271 | 229 |
+| FloatingIslands | ThicknessRatio | 2 | 0.100000001 | 0.699999988 | 0.399999995 | 0.299999993 |
+| FloatingIslands | VerticalJitter | 2 | 0.200000003 | 0.600000024 | 0.400000013 | 0.20000001 |
+| FloatingIslands | TopFlatten | 2 | 0.600000024 | 0.800000012 | 0.700000018 | 0.099999994 |
+| FloatingIslands | SurfaceRoughness | 2 | 0 | 4 | 2 | 2 |

 The reflection/use audit found **36 excluded fields**: the 34 terrain-op transport fields
 `TerraceStepHeight`, `TerraceHardness`, `TerraceNoiseDisplacement`, `LayerLineSpacing`,
@@ -222,22 +270,45 @@ The reflection/use audit found **36 excluded fields**: the 34 terrain-op transpo
 `ChimneyMaxRadius`, `ChimneyHeight`, `DomeDensity`, `DomeMinRadius`, `DomeMaxRadius`,
 `DomeHeightRatio`, `PinchDensity`, `PinchStrength`, `PinchLength` — populated per room by
 `UVoxelTerrainOpDefinition` — plus the manager-owned runtime bounds `StrateTopWorldZ` and
-`StrateBottomWorldZ`. The test prints the complete field-by-field table, including excluded fields,
+`StrateBottomWorldZ`. The tunnel transport fields are measured directly from the native
+`FStrateGenerationParams` X-macro even though they are not reflected `UPROPERTY`s; their values are
+reported but deliberately not jittered. Same-named fields in sibling structs are not automatically
+excluded: the exclusion applies to the tunnel transport slots whose generation source is the
+terrain-op pool. The test prints the complete field-by-field table, including excluded fields,
 reflection status, and clamp metadata.

+The terrain-field source is now traced in code. `VoxelStrateManager.cpp` loads the active strate's
+`TerrainOperations` soft-object pool, and `VoxelGenerator.cpp` passes that pool to
+`VoxelCaveMorphology::BuildChunkCache`. For each nearest room, `FVoxelTerrainOpDefinition::ApplyTo`
+overlays the selected operation onto a local copy (`FStrateGenerationParams LocalTerrainParams =
+Params`) before the room operator runs. Thus the 34 values are transported through the strate
+params, but their authored cave-detail values come from `UVoxelTerrainOpDefinition` assets selected
+by the active strate's terrain-operation pool. Rolling those slots in the strate composer is not a
+useful way to roll cave detail: selected room operations overwrite them. It is not a literal
+absolute no-op, because the local base copy is still used where no operation overwrites a slot.
+The correct future design is a **second corpus** over terrain-op assets plus the active pool's
+membership/weights; this task does not build it. The project currently has **1**
+`UVoxelTerrainOpDefinition` asset (the Asset Registry count used by the test).
+
 The bool policy is explicit: `bTunnelsFlowTowardOrigin` is inherited from the dominant parent.
 `OriginRoomMaxConnections` and `RoughnessNoiseType` retain `Lerp`'s existing SNAP behavior and
 are not jittered. In this editor build clamp metadata was available while the table was built; it
 is **not promised in a cooked runtime**, so the commandlet/cook-time bake remains owed.

-The first 64-candidate run (step 4, radius 256, max 8,000,000 cells, fixture seed 1337) produced
-64/64 non-vacuous candidates, 64/64 largest-component share ≥ 0.50, and 64/64 exact unsnapped
-arrival→departure law passes. Every row was `air=0.520272`, `largest share=0.999996`,
-`walkable=0.057995`, `feature scale=20.000000`, `Connected`; total runtime was **542.220 s**.
-The §6.2 check saw **2,560 Mixed, 0 AllSolid, 0 AllAir, 0 brute-force voxels, 0 violations**:
-this is explicitly **vacuous**, not evidence that rolled bounds are sound, because the current cave
-box path emitted no uniform verdict. The test leaves §6.2 open and reports that fact rather than
-turning it into a workaround.
+The fed 64-candidate run (step 4 sweep, radius 256, max 8,000,000 cells, fixture seed 1337)
+produced **59/64 survivors (92.2%)**. There were 60 non-vacuous candidates, 60 with largest
+component share ≥ 0.50, 63 exact unsnapped arrival→departure law passes, and no roll failures.
+The complete 64-row table is emitted by `VoxelForge.Composer.ParameterRoll`; each row reports
+archetype, weighted parents, air fraction, largest share, walkable fraction, feature scale, and
+arrival→departure verdict. The final full-suite run took **112.574 s** total (**0.015 s** corpus
+load, **112.556 s** roll/measurement).
+
+The §6.2 check is no longer vacuous: **1,224 boxes proved**, **1,629,144 voxels checked**,
+**0 violations**. Its verdict mix was 1,336 Mixed, 549 AllSolid, and 675 AllAir; the law check
+reported no violating candidate. The box law used unit step and brute-forced every voxel of each
+uniform box, while the candidate sweep retained step 4. The test asserts zero violations and keeps
+the law open to future changed candidates rather than treating this clean run as a license to tune
+the survival rate.

 ### 3.4 The measurement pass — one grid, one flood fill, three jobs

@@ -4,10 +4,15 @@

 #include "Misc/AutomationTest.h"
 #include "HAL/PlatformTime.h"
+#include "AssetRegistry/AssetData.h"
+#include "AssetRegistry/AssetRegistryModule.h"
+#include "Modules/ModuleManager.h"

 #include "VoxelForgeTestFixture.h"
+#include "VoxelSettings.h"
 #include "VoxelStrateComposer.h"
 #include "VoxelStrateMeasure.h"
+#include "VoxelTerrainOpDefinition.h"

 namespace
 {
@@ -53,6 +58,124 @@ namespace
                                *VF_FormatSpreadValue(Spread.ClampMax));
    }

+    bool VF_IsProjectPackagePath(const FString& PackagePath)
+    {
+        return PackagePath == TEXT("/Game")
+            || PackagePath.StartsWith(TEXT("/Game/"), ESearchCase::IgnoreCase);
+    }
+
+    bool VF_IsIgnoredSavedCopy(const FString& PackagePath)
+    {
+        return PackagePath.Contains(TEXT("/Saved/Autosaves"), ESearchCase::IgnoreCase)
+            || PackagePath.Contains(TEXT("/Saved/Cooked"), ESearchCase::IgnoreCase);
+    }
+
+    TArray<FString> VF_GetProjectAssetPaths(UClass* AssetClass)
+    {
+        TArray<FString> Paths;
+        if (AssetClass == nullptr)
+        {
+            return Paths;
+        }
+
+        FAssetRegistryModule& AssetRegistryModule =
+            FModuleManager::LoadModuleChecked<FAssetRegistryModule>(TEXT("AssetRegistry"));
+        TArray<FAssetData> Assets;
+        AssetRegistryModule.Get().GetAssetsByClass(AssetClass->GetClassPathName(), Assets, true);
+
+        TSet<FString> SeenPaths;
+        for (const FAssetData& Asset : Assets)
+        {
+            const FString PackagePath = Asset.PackageName.ToString();
+            const FString ObjectPath = Asset.GetObjectPathString();
+            if (!VF_IsProjectPackagePath(PackagePath)
+                || VF_IsIgnoredSavedCopy(PackagePath)
+                || VF_IsIgnoredSavedCopy(ObjectPath)
+                || SeenPaths.Contains(ObjectPath))
+            {
+                continue;
+            }
+            SeenPaths.Add(ObjectPath);
+            Paths.Add(ObjectPath);
+        }
+        Paths.Sort();
+        return Paths;
+    }
+
+    FString VF_FormatSettingsAudit(const UVoxelSettings* Settings)
+    {
+        TArray<FString> Paths;
+        if (Settings != nullptr)
+        {
+            TSet<FString> UniquePaths;
+            for (const TPair<int32, TSoftObjectPtr<UVoxelStrateDefinition>>& Pair : Settings->FixedStrates)
+            {
+                const FString Path = Pair.Value.ToSoftObjectPath().ToString();
+                if (!Path.IsEmpty())
+                {
+                    UniquePaths.Add(Path);
+                }
+            }
+            for (const TSoftObjectPtr<UVoxelStrateDefinition>& Reference : Settings->StratePool)
+            {
+                const FString Path = Reference.ToSoftObjectPath().ToString();
+                if (!Path.IsEmpty())
+                {
+                    UniquePaths.Add(Path);
+                }
+            }
+            for (const FString& Path : UniquePaths)
+            {
+                Paths.Add(Path);
+            }
+            Paths.Sort();
+        }
+
+        FString Result = FString::Printf(TEXT("Settings corpus audit: %d unique fixed/pool paths"),
+                                          Paths.Num());
+        for (const FString& Path : Paths)
+        {
+            Result += FString::Printf(TEXT(" [%s]"), *Path);
+        }
+        Result += TEXT("; these settings references are not the corpus source.");
+        return Result;
+    }
+
+    bool VF_RollRuntimeZIsZero(const FVoxelStrateRollInfo& Roll)
+    {
+        switch (Roll.Archetype)
+        {
+        case ECaveGeneratorType::TunnelNetwork:
+        case ECaveGeneratorType::Underwater:
+            return Roll.ArchetypeParams.TunnelNetworkParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.TunnelNetworkParams.StrateBottomWorldZ == 0.0f;
+        case ECaveGeneratorType::FlatPlain:
+        case ECaveGeneratorType::CrystalChamber:
+            return Roll.ArchetypeParams.SlabParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.SlabParams.StrateBottomWorldZ == 0.0f;
+        case ECaveGeneratorType::Maze:
+            return Roll.ArchetypeParams.MazeParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.MazeParams.StrateBottomWorldZ == 0.0f;
+        case ECaveGeneratorType::SurfaceWorld:
+            return Roll.ArchetypeParams.SurfaceParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.SurfaceParams.StrateBottomWorldZ == 0.0f;
+        case ECaveGeneratorType::VerticalShafts:
+            return Roll.ArchetypeParams.VerticalShaftParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.VerticalShaftParams.StrateBottomWorldZ == 0.0f;
+        case ECaveGeneratorType::FloatingIslands:
+            return Roll.ArchetypeParams.FloatingIslandParams.StrateTopWorldZ == 0.0f
+                && Roll.ArchetypeParams.FloatingIslandParams.StrateBottomWorldZ == 0.0f;
+        default:
+            return false;
+        }
+    }
+
+    bool VF_IsComposerDefault(const FVoxelStrateCorpusEntry& Entry)
+    {
+        return Entry.SourcePath.StartsWith(TEXT("/VoxelForge/ComposerDefaults/"),
+                                           ESearchCase::IgnoreCase);
+    }
+
    FString VF_FormatParents(const FVoxelStrateCorpus& Corpus,
                             const FVoxelStrateRollInfo& Roll)
    {
@@ -233,21 +356,81 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
    FVoxelStrateCorpus Corpus;
    FString CorpusReport;
    const double CorpusStartSeconds = FPlatformTime::Seconds();
-    const bool bCorpusLoaded = Corpus.LoadFromSettings(AuthoredSettings, CorpusReport);
+    const bool bCorpusLoaded = Corpus.LoadFromAssetRegistry(CorpusReport);
    const double CorpusSeconds = FPlatformTime::Seconds() - CorpusStartSeconds;
    AddInfo(FString::Printf(TEXT("%s Load time %.3fs; contents hash 0x%08x; clamp metadata at "
                                 "build=%s, promised at cooked runtime=%s."),
                            *CorpusReport, CorpusSeconds, Corpus.GetContentsHash(),
                            Corpus.AreClampMetadataAvailableAtBuild() ? TEXT("yes") : TEXT("no"),
                            Corpus.AreClampMetadataAvailableAtRuntime() ? TEXT("yes") : TEXT("no")));
+    AddInfo(VF_FormatSettingsAudit(AuthoredSettings));
+
+    const TArray<FString> ProjectStrateAssets =
+        VF_GetProjectAssetPaths(UVoxelStrateDefinition::StaticClass());
+    const TArray<FString> ProjectTerrainOpAssets =
+        VF_GetProjectAssetPaths(UVoxelTerrainOpDefinition::StaticClass());
+    int32 NumProjectCorpusEntries = 0;
+    int32 NumDefaultCorpusEntries = 0;
+    FString CorpusMembership = TEXT("CORPUS MEMBERSHIP\n");
+    for (const FVoxelStrateCorpusEntry& Entry : Corpus.GetEntries())
+    {
+        if (VF_IsComposerDefault(Entry))
+        {
+            ++NumDefaultCorpusEntries;
+        }
+        else
+        {
+            ++NumProjectCorpusEntries;
+        }
+        CorpusMembership += FString::Printf(TEXT("%s | %s | %s\n"),
+                                             VF_IsComposerDefault(Entry) ? TEXT("default") : TEXT("project"),
+                                             VF_GetStrateArchetypeName(Entry.Archetype),
+                                             *Entry.SourcePath);
+    }
+    AddInfo(FString::Printf(
+        TEXT("Corpus membership: %d project strate assets discovered, %d project vectors loaded, "
+             "%d defaults, %d total members. Terrain-op Asset Registry count (not rolled in this "
+             "task): %d."),
+        ProjectStrateAssets.Num(), NumProjectCorpusEntries, NumDefaultCorpusEntries, Corpus.Num(),
+        ProjectTerrainOpAssets.Num()));
+    AddInfo(CorpusMembership);
+    for (const ECaveGeneratorType Archetype : {
+        ECaveGeneratorType::TunnelNetwork,
+        ECaveGeneratorType::FlatPlain,
+        ECaveGeneratorType::CrystalChamber,
+        ECaveGeneratorType::Maze,
+        ECaveGeneratorType::SurfaceWorld,
+        ECaveGeneratorType::VerticalShafts,
+        ECaveGeneratorType::FloatingIslands,
+        ECaveGeneratorType::Underwater})
+    {
+        int32 NumDefaultsForArchetype = 0;
+        for (const FVoxelStrateCorpusEntry& Entry : Corpus.GetEntries())
+        {
+            if (Entry.Archetype == Archetype && VF_IsComposerDefault(Entry))
+            {
+                ++NumDefaultsForArchetype;
+            }
+        }
+        TestEqual(FString::Printf(TEXT("one hand-tuned default for %s"),
+                                  VF_GetStrateArchetypeName(Archetype)),
+                  NumDefaultsForArchetype, 1);
+        TestTrue(FString::Printf(TEXT("%s has a same-archetype corpus group"),
+                                 VF_GetStrateArchetypeName(Archetype)),
+                 Corpus.NumForArchetype(Archetype) > 0);
+    }
+    TestEqual(TEXT("all project strate assets plus eight defaults are corpus members"),
+              Corpus.Num(), ProjectStrateAssets.Num() + 8);

    FString SpreadTable = TEXT(
        "CORPUS SPREAD (population stddev; jitter range = ±15% of max-min; clamps are safety only)\n"
-        "field | kind | excluded | reflected | samples | min | max | mean | stddev | clamp\n");
+        "archetype | struct | field | kind | excluded | reflected | samples | min | max | mean | stddev | clamp\n");
    for (const FVoxelStrateFieldSpread& Spread : Corpus.GetFieldSpreads())
    {
        SpreadTable += FString::Printf(
-            TEXT("%s | %s | %s | %s | %d | %s | %s | %s | %s | %s\n"),
+            TEXT("%s | %s | %s | %s | %s | %s | %d | %s | %s | %s | %s | %s\n"),
+            VF_GetStrateArchetypeName(Spread.Archetype),
+            *Spread.ParamStructName,
            *Spread.FieldName,
            VF_FieldKindName(Spread.Kind),
            Spread.bExcluded ? TEXT("yes") : TEXT("no"),
@@ -270,9 +453,10 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
    AddInfo(ExclusionTable);
    AddInfo(TEXT(
        "Bool policy: bTunnelsFlowTowardOrigin is inherited from the dominant weighted parent; "
-        "int/enum SNAP fields use FStrateGenerationParams::Lerp and are not jittered. Parent "
-        "selection is weight-1.0 (equal) because UVoxelStrateDefinition has no corpus-weight field; "
-        "parents are selected only within one archetype group."));
+        "int/enum SNAP fields use the native family blend and are not jittered. Parent selection "
+        "is weight-1.0 (equal) because UVoxelStrateDefinition has no corpus-weight field; parents "
+        "are selected only within one exact archetype group. FStrateGenerationParams terrain-op "
+        "transport fields remain excluded; the sibling families' authored fields are rolled."));

    TestTrue(TEXT("the reflected parameter schema is covered or explicitly excluded"),
             Corpus.IsSchemaValid());
@@ -281,7 +465,7 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
    {
        AddError(FString::Printf(
            TEXT("Cannot run the 64-candidate measurement: %s"),
-            Corpus.IsSchemaValid() ? TEXT("the settings asset exposed no TunnelNetwork/Underwater GenerationParams")
+            Corpus.IsSchemaValid() ? TEXT("the Asset Registry exposed no usable strate vectors")
                                   : *Corpus.GetSchemaError()));
        return false;
    }
@@ -326,6 +510,7 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
    int64 TotalBoxCheckedVoxels = 0;
    int32 TotalBoxViolations = 0;
    TArray<int32> BoxViolationCandidates;
+    FString FirstBoxViolation;

    const double RollAndMeasureStartSeconds = FPlatformTime::Seconds();
    for (int32 CandidateIndex = 0; CandidateIndex < NumCandidates; ++CandidateIndex)
@@ -349,23 +534,47 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
            continue;
        }

+        bool bParentsMatchArchetype = true;
+        for (const int32 ParentIndex : Roll.ParentEntryIndices)
+        {
+            if (!Corpus.GetEntries().IsValidIndex(ParentIndex)
+                || Corpus.GetEntries()[ParentIndex].Archetype != Roll.Archetype)
+            {
+                bParentsMatchArchetype = false;
+                break;
+            }
+        }
+        TestTrue(FString::Printf(TEXT("candidate %d parents stay within %s"),
+                                 CandidateIndex, VF_GetStrateArchetypeName(Roll.Archetype)),
+                 bParentsMatchArchetype);
+
+        const FVoxelStrateRollInfo RepeatRoll = VF_RollStrateParamsDetailed(
+            Corpus, AuthoredSettings->Seed, CandidateIndex);
+        TestTrue(FString::Printf(TEXT("candidate %d is bit-identical on deterministic reroll"),
+                                 CandidateIndex),
+                 RepeatRoll.bValid && RepeatRoll.Archetype == Roll.Archetype
+                     && VF_AreStrateArchetypeParamsBitIdentical(
+                         RepeatRoll.ArchetypeParams, Roll.ArchetypeParams, Roll.Archetype)
+                     && RepeatRoll.ParentEntryIndices == Roll.ParentEntryIndices
+                     && RepeatRoll.ParentWeights == Roll.ParentWeights);
+
        // Put the invented vector into a transient fixture definition. This is an offline test
        // world only; the runtime manager/generation code and authored assets are untouched.
        UVoxelStrateDefinition* CandidateDefinition =
            World.Definitions[CandidateStrateIndex].Get();
        CandidateDefinition->GeneratorType = Roll.Archetype;
-        CandidateDefinition->GenerationParams = Roll.Params;
+        CandidateDefinition->GenerationParams = Roll.ArchetypeParams.TunnelNetworkParams;
+        CandidateDefinition->SlabParams = Roll.ArchetypeParams.SlabParams;
+        CandidateDefinition->MazeParams = Roll.ArchetypeParams.MazeParams;
+        CandidateDefinition->SurfaceParams = Roll.ArchetypeParams.SurfaceParams;
+        CandidateDefinition->VerticalShaftParams = Roll.ArchetypeParams.VerticalShaftParams;
+        CandidateDefinition->FloatingIslandParams = Roll.ArchetypeParams.FloatingIslandParams;
        CandidateDefinition->bUseOperatorStack = true;
        CandidateDefinition->TransitionType = EVoxelStrateTransition::Hard;
        World.Reinitialize();

-        const FStrateGenerationParams RuntimeParams = World.StrateManager->GetGenerationParams(
-            FIntVector(0, 0,
-                World.StrateManager->GetLayout()[CandidateStrateIndex].BottomChunkZ
-                + (World.StrateManager->GetLayout()[CandidateStrateIndex].TopChunkZ
-                   - World.StrateManager->GetLayout()[CandidateStrateIndex].BottomChunkZ) / 2));
        TestTrue(FString::Printf(TEXT("candidate %d keeps runtime Z out of the rolled vector"), CandidateIndex),
-                 Roll.Params.StrateTopWorldZ == 0.0f && Roll.Params.StrateBottomWorldZ == 0.0f);
+                 VF_RollRuntimeZIsZero(Roll));

        const FVoxelStrateMetrics Metrics = VF_MeasureStrate(
            *World.Generator, *World.StrateManager, CandidateStrateIndex, MeasureSettings);
@@ -438,6 +647,10 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
        if (BoxReport.Violations > 0)
        {
            BoxViolationCandidates.Add(CandidateIndex);
+            if (FirstBoxViolation.IsEmpty())
+            {
+                FirstBoxViolation = BoxReport.FirstViolation;
+            }
        }

        CandidateTable += FString::Printf(
@@ -481,8 +694,8 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
        AddError(FString::Printf(
            TEXT("ROLLED BOX-VERDICT VIOLATION: %d candidate(s) produced a false AllSolid/AllAir "
                 "claim; first candidate index %d. This closes the test red because §6.2 bounds "
-                 "were not sound for the rolled parameter space."),
-            BoxViolationCandidates.Num(), BoxViolationCandidates[0]));
+                 "were not sound for the rolled parameter space. First violation: %s"),
+            BoxViolationCandidates.Num(), BoxViolationCandidates[0], *FirstBoxViolation));
    }
    if (TotalBoxProved == 0)
    {
@@ -504,4 +717,3 @@ bool FVoxelForgeComposerParameterRollTest::RunTest(const FString& Parameters)
 }

 #endif // WITH_DEV_AUTOMATION_TESTS
-
@@ -28,9 +28,11 @@ struct VOXELFORGE_API FVoxelStrateFieldExclusion
    FString Reason;
 };

-/** Measured statistics for one field across the currently loaded corpus. */
+/** Measured statistics for one field in one archetype's corpus group. */
 struct VOXELFORGE_API FVoxelStrateFieldSpread
 {
+    ECaveGeneratorType Archetype = ECaveGeneratorType::TunnelNetwork;
+    FString ParamStructName;
    FString FieldName;
    EVoxelStrateFieldKind Kind = EVoxelStrateFieldKind::Continuous;
    bool bExcluded = false;
@@ -50,12 +52,30 @@ struct VOXELFORGE_API FVoxelStrateFieldSpread
    double ClampMax = 0.0;
 };

+/** Native parameter storage for every currently supported strate archetype. */
+struct VOXELFORGE_API FVoxelStrateArchetypeParams
+{
+    FStrateGenerationParams TunnelNetworkParams;
+    FSlabGenerationParams SlabParams;
+    FMazeGenerationParams MazeParams;
+    FSurfaceGenerationParams SurfaceParams;
+    FVerticalShaftParams VerticalShaftParams;
+    FFloatingIslandParams FloatingIslandParams;
+};
+
 /** One known-good authored vector and the archetype that gives it meaning. */
 struct VOXELFORGE_API FVoxelStrateCorpusEntry
 {
    FString SourcePath;
    FString SourceName;
    ECaveGeneratorType Archetype = ECaveGeneratorType::TunnelNetwork;
+
+    // Native storage is selected by Archetype. Keeping every family here makes the corpus
+    // type-safe without changing any runtime generation parameter struct.
+    FVoxelStrateArchetypeParams ArchetypeParams;
+
+    // Compatibility view for callers of the original Tier 4a API. It is meaningful for
+    // TunnelNetwork/Underwater entries and is the same value as ArchetypeParams.TunnelNetworkParams.
    FStrateGenerationParams Params;

    // The current asset format has no corpus weight field. Loaded hand-authored entries therefore
@@ -71,6 +91,10 @@ struct VOXELFORGE_API FVoxelStrateRollInfo
    FString FailureReason;

    ECaveGeneratorType Archetype = ECaveGeneratorType::TunnelNetwork;
+    FVoxelStrateArchetypeParams ArchetypeParams;
+
+    // Compatibility view for the original tunnel-only API. For a non-tunnel roll, the native
+    // result is in ArchetypeParams and this member is not the active candidate vector.
    FStrateGenerationParams Params;

    TArray<int32> ParentEntryIndices;
@@ -81,17 +105,22 @@ struct VOXELFORGE_API FVoxelStrateRollInfo
 /**
 * Known-good strate vectors plus their measured field spread.
 *
- * `LoadFromSettings` walks both the pool and fixed-strate references, resolves each definition,
- * de-duplicates it by asset path, and admits only TunnelNetwork/Underwater definitions because
- * those are the two archetypes whose authored vector is `GenerationParams`. Other archetypes have
- * different parameter structs; treating their default `GenerationParams` as authored data would
- * manufacture a false corpus.
+ * `LoadFromAssetRegistry` enumerates every project UVoxelStrateDefinition through the Asset
+ * Registry, explicitly excluding Saved/Autosaves and Saved/Cooked copies. It also adds one
+ * known-good default vector for every archetype/family in VoxelStrateTypes.h. Parents are selected
+ * only inside the exact archetype group; no cross-archetype blend is attempted.
 */
 class VOXELFORGE_API FVoxelStrateCorpus
 {
 public:
    void Reset();

+    bool LoadFromAssetRegistry(FString& OutReport);
+
+    /**
+     * Compatibility alias retained for callers compiled against the first Tier 4a pass. The
+     * settings argument is audited only for diagnostics; it is not a corpus input.
+     */
    bool LoadFromSettings(const UVoxelSettings* Settings, FString& OutReport);
    bool LoadFromDefinitions(const TArray<UVoxelStrateDefinition*>& Definitions, FString& OutReport);

@@ -99,6 +128,9 @@ public:
    bool AddEntry(const FString& SourcePath, const FString& SourceName,
                  ECaveGeneratorType Archetype, const FStrateGenerationParams& Params,
                  float Weight = 1.0f);
+    bool AddEntry(const FString& SourcePath, const FString& SourceName,
+                  ECaveGeneratorType Archetype, const FVoxelStrateArchetypeParams& Params,
+                  float Weight = 1.0f);

    bool IsValid() const { return bSchemaValid && Entries.Num() > 0; }
    bool IsSchemaValid() const { return bSchemaValid; }
@@ -109,6 +141,10 @@ public:
    const TArray<FVoxelStrateFieldSpread>& GetFieldSpreads() const { return FieldSpreads; }
    const TArray<FString>& GetSkippedDefinitions() const { return SkippedDefinitions; }

+    int32 NumForArchetype(ECaveGeneratorType Archetype) const;
+
+    const FVoxelStrateFieldSpread* FindSpread(
+        ECaveGeneratorType Archetype, const FString& FieldName) const;
    const FVoxelStrateFieldSpread* FindSpread(const FString& FieldName) const;

    /** Stable hash of source paths, archetypes, weights, and every listed scalar field. */
@@ -149,3 +185,8 @@ VOXELFORGE_API FStrateGenerationParams VF_RollStrateParams(
 VOXELFORGE_API bool VF_AreStrateParamsBitIdentical(
    const FStrateGenerationParams& A, const FStrateGenerationParams& B);

+/** Bitwise comparison of the native vector for one archetype family. */
+VOXELFORGE_API bool VF_AreStrateArchetypeParamsBitIdentical(
+    const FVoxelStrateArchetypeParams& A,
+    const FVoxelStrateArchetypeParams& B,
+    ECaveGeneratorType Archetype);