Merge pull request 'fix(M00): release GUI capture render resources' (#15) from work/sindo-main-codex/m00-gui-rid-shutdown into master

Reviewed-on: #15

RENDER.md

@@ -822,6 +822,14 @@ $exe = Join-Path $env:TEMP 'godot-4.6.1-openwc\Godot_v4.6.1-stable_win64.exe'
 - Added `src/tools/capture_render_checkpoints.gd` to load the Eastern Kingdoms streaming scene, move a dedicated camera to fixed regression viewpoints and save PNGs under `user://render_checkpoints`.
 - Current checkpoints cover the Elwynn waterfall from front/side and Goldshire Inn/Blacksmith windows. Headless runs are dry-run only because the headless display server cannot produce viewport PNGs.
 
+## 2026-07-13 Renderer Shutdown Ownership
+
+- `StreamingWorldLoader` drains its asynchronous work before teardown, leaves parented streamed nodes to SceneTree ownership, then explicitly frees detached M2/WMO prototype nodes and clears resource caches.
+- `ADTBuilder` explicitly frees a temporary `Water` root when an ADT liquid payload produces no child geometry. Dry tiles must not leave empty nodes in ObjectDB.
+- `src/tools/verify_render_runtime_cache_shutdown.gd` regression-checks detached prototype and resource-cache release plus empty-liquid-root ownership. The unified render-baseline runner executes it before capture.
+- A focused 8-second GUI checkpoint and a seven-checkpoint single-pass GUI run finish without Node, renderer resource or RID leak diagnostics. A timing-dependent anonymous zero-reference `RefCounted` ObjectDB warning can remain at engine teardown; it is tracked separately and is not suppressed.
+- This is shutdown-only ownership behavior and does not alter checkpoint rendering, cache formats or the `Blizzlike335` compatibility profile.
+
 ## Practical Rule For Future Work
 
 If something improves quality but creates visible hitch, it is not done. Move it to bake/cache/background work, split finalization over frames, or prewarm it before the player can see it.

addons/mpq_extractor/loaders/adt_builder.gd

@@ -788,7 +788,10 @@ static func _build_tile_water_root(data: Dictionary, origin_offset: Vector3) ->
 		mi.cast_shadow = GeometryInstance3D.SHADOW_CASTING_SETTING_OFF
 		root.add_child(mi)
 
-	return root if root.get_child_count() > 0 else null
+	if root.get_child_count() > 0:
+		return root
+	root.free()
+	return null
 
 
 static func _append_liquid_geometry(

coordination/claims/M00-QAR-ANIMATION-SKY-EVIDENCE-001.md

@@ -0,0 +1,78 @@
+# M00-QAR-ANIMATION-SKY-EVIDENCE-001 — Animation and sky evidence
+
+<!-- OPENWC_CLAIM:M00-QAR-ANIMATION-SKY-EVIDENCE-001:sindo-main-codex:2026-07-15 -->
+
+## Ownership
+
+- Target: M00
+- Program: QAR
+- Owner/Agent ID: sindo-main-codex
+- Branch: `work/sindo-main-codex/m00-animation-sky-evidence`
+- Lease expires UTC: 2026-07-15
+- Integrator: milestone integrator
+
+## Outcome
+
+Add private build 12340 evidence for the remaining sky-transition and native animated-M2 coverage, rerun all paired comparisons, and hand the resulting audit to the milestone integrator.
+
+## Non-goals
+
+- Committing proprietary screenshots or extracted assets.
+- Claiming renderer parity from mismatched world placement or animation phase.
+- Marking M00 DONE; only the milestone integrator may change target status or Evidence.
+
+## Paths
+
+- Exclusive: this claim and animation/sky evidence documentation
+- Shared/hotspots: `docs/M00_FINAL_PAIRED_EVIDENCE.md`, `docs/modules/world-renderer.md`
+- Generated/ignored: normalized private references under `sources/OpenWCReferenceCheckpoints`; captures and reports under `user://`
+
+## Contracts and data
+
+- Public API/events: unchanged
+- Schema/format version: unchanged
+- Migration/compatibility: existing five-reference comparison remains valid
+- Consumers: M00 milestone integrator
+
+## Dependencies
+
+- Requires: user-captured build 12340 screenshots at 19:00 and two real `GryphonRoost01` animation phases
+- Blocks: final integrator decision on M00 completion
+- External state: private screenshots remain outside Git
+
+## Verification
+
+- Commands: private-reference preparation, full baseline runner, comparator self-test, coordination and documentation gates
+- Fixtures: seven private build 12340 checkpoint references plus a second animation-phase provenance frame
+- Fidelity evidence: sky-transition screenshot and observed wing pose change over three seconds
+- Performance budget: existing M00 capture budgets
+
+## Documentation deliverables
+
+- Inline public API docs: unchanged
+- Module specification: paired coverage and remaining limitations
+- Data-flow diagram: unchanged
+- Sequence/state/dependency diagrams: unchanged
+- Source map/status updates: final paired evidence audit
+
+## Simplicity and naming
+
+- Important names introduced: none
+- Simplest considered solution: normalize private captures outside Git and reuse the existing comparator
+- Rejected complexity/abstractions: adding a video comparator or committing an asset corpus
+- Unavoidable complexity and justification: animation phase and synthetic placement require human evidence classification alongside pixel metrics
+- Measured optimization evidence: not applicable
+
+## Status
+
+- State: ready-for-review
+- Done: exact-size private references registered; real ADT placement verified; canonical and visual captures completed; fourteen comparisons audited
+- Next: milestone integrator reviews synthetic/real placement classification and GUI RID risk
+- Blocked by:
+
+## Handoff
+
+- Commit: this work-package commit
+- Results: `compared=14`, `failed=14`, `missing=0`; mean error `0.066893..0.176955`; original-client Gryphon wing phase changed over three seconds
+- Remaining risks: synthetic probe placement differs from the original-client world placement; GUI RID/resource diagnostics remain separate
+- Documentation updated: `docs/M00_FINAL_PAIRED_EVIDENCE.md`, `docs/modules/world-renderer.md`

coordination/claims/M00-QAR-FINAL-PAIRED-EVIDENCE-001.md

@@ -0,0 +1,78 @@
+# M00-QAR-FINAL-PAIRED-EVIDENCE-001 — Final paired evidence audit
+
+<!-- OPENWC_CLAIM:M00-QAR-FINAL-PAIRED-EVIDENCE-001:sindo-main-codex:2026-07-14 -->
+
+## Ownership
+
+- Target: M00
+- Program: QAR
+- Owner/Agent ID: sindo-main-codex
+- Branch: `work/sindo-main-codex/m00-final-paired-evidence`
+- Lease expires UTC: 2026-07-14
+- Integrator: milestone integrator
+
+## Outcome
+
+Produce one final WMO-ready paired report for all available build 12340 references and reconcile the remaining M00 acceptance gaps.
+
+## Non-goals
+
+- Achieving renderer parity inside M00.
+- Calibrating unknown manual camera parameters.
+- Marking the target DONE; only the milestone integrator may do so.
+
+## Paths
+
+- Exclusive: paired-run exit handling and final evidence record
+- Shared/hotspots: renderer baseline/module findings
+- Generated/ignored: final captures and reports under `user://`
+
+## Contracts and data
+
+- Public API/events: visual comparison exit 0/1 both count as completed report generation
+- Schema/format version: unchanged
+- Migration/compatibility: comparator infrastructure errors remain fatal
+- Consumers: M00 integrator
+
+## Dependencies
+
+- Requires: merged region diagnostic and five private reference images
+- Blocks: integrator decision on M00 completion
+- External state: private images remain outside Git
+
+## Verification
+
+- Commands: complete paired baseline run, report audit, repository gates
+- Fixtures: five build 12340 references
+- Fidelity evidence: cold/warm comparison metrics and human gap classification
+- Performance budget: existing M00 capture budgets
+
+## Documentation deliverables
+
+- Inline public API docs: PowerShell comparison step semantics
+- Module specification: final evidence status
+- Data-flow diagram: unchanged
+- Sequence/state/dependency diagrams: unchanged
+- Source map/status updates: baseline acceptance summary
+
+## Simplicity and naming
+
+- Important names introduced: `Invoke-GodotComparisonStep`
+- Simplest considered solution: allow only comparator exit codes 0 and 1
+- Rejected complexity/abstractions: suppressing all process failures
+- Unavoidable complexity and justification: threshold exceedance is evidence, not infrastructure failure
+- Measured optimization evidence: not applicable
+
+## Status
+
+- State: ready-for-review
+- Done: paired-run exit contract fixed; full baseline and visual capture completed; ten pairs audited; final acceptance document written
+- Next: integrator decides animated/sky reference policy and GUI RID-risk disposition
+- Blocked by:
+
+## Handoff
+
+- Commit: this work-package commit
+- Results: compared=10, failed=10, missing=0; mean error 0.066892..0.176952; cold/warm gaps stable; full workflow returned success with threshold evidence preserved
+- Remaining risks: no build 12340 animated-M2 or sky-transition reference; GUI renderer RID/resources still reported at exit
+- Documentation updated: `docs/M00_FINAL_PAIRED_EVIDENCE.md`, `docs/modules/world-renderer.md`

coordination/claims/M00-QAR-GUI-RID-SHUTDOWN-001.md

@@ -0,0 +1,78 @@
+# M00-QAR-GUI-RID-SHUTDOWN-001 — GUI capture RID shutdown
+
+<!-- OPENWC_CLAIM:M00-QAR-GUI-RID-SHUTDOWN-001:sindo-main-codex:2026-07-15 -->
+
+## Ownership
+
+- Target: M00
+- Program: QAR
+- Owner/Agent ID: sindo-main-codex
+- Branch: `work/sindo-main-codex/m00-gui-rid-shutdown`
+- Lease expires UTC: 2026-07-15
+- Integrator: milestone integrator
+
+## Outcome
+
+Identify ownership of GUI render-baseline shutdown leaks and apply the smallest regression-covered fix that lets queued world resources release before engine teardown.
+
+## Non-goals
+
+- Renderer architecture decomposition.
+- Hiding or filtering Godot leak diagnostics.
+- Changing M00 status, Evidence or `OPENWC_TARGET_DONE`.
+
+## Paths
+
+- Exclusive: capture shutdown sequencing and its regression diagnostic
+- Shared/hotspots: `src/tools/capture_render_checkpoints.gd`, renderer module shutdown documentation
+- Generated/ignored: GUI capture output and verbose logs under `user://`
+
+## Contracts and data
+
+- Public API/events: unchanged
+- Schema/format version: unchanged
+- Migration/compatibility: capture images and report schema remain unchanged
+- Consumers: M00 baseline runner and milestone integrator
+
+## Dependencies
+
+- Requires: merged seven-reference evidence audit
+- Blocks: integrator disposition of the final GUI renderer leak risk
+- External state: proprietary references are not required for the focused shutdown regression
+
+## Verification
+
+- Commands: focused GUI single-checkpoint capture with verbose shutdown, headless project load, manifest/self-tests, repository gates
+- Fixtures: existing non-proprietary render-baseline manifest and local extracted corpus
+- Fidelity evidence: no visual behavior change; focused PNG hash/report remains valid
+- Performance budget: shutdown-only change
+
+## Documentation deliverables
+
+- Inline public API docs: shutdown method semantics if introduced
+- Module specification: shutdown ownership and observed leak status
+- Data-flow diagram: unchanged
+- Sequence/state/dependency diagrams: update shutdown sequence only if behavior changes
+- Source map/status updates: capture tool if changed
+
+## Simplicity and naming
+
+- Important names introduced: `_release_runtime_caches_for_shutdown`, `_free_detached_node_cache`, `_queue_free_streamed_node`
+- Simplest considered solution: explicitly release capture-owned scene references and drain deferred deletion
+- Rejected complexity/abstractions: global RID registry or suppressed engine diagnostics
+- Unavoidable complexity and justification: async streaming jobs must finish before resource ownership can drain
+- Measured optimization evidence: not applicable
+
+## Status
+
+- State: ready-for-review
+- Done: reproduced the focused leak; assigned detached prototype/resource caches and empty liquid roots to explicit owners; added a headless regression; verified zero Node/resource/RID leak lines in focused and seven-checkpoint GUI captures
+- Next: integrator review and disposition of the residual engine-level zero-reference `RefCounted` diagnostic
+- Blocked by:
+
+## Handoff
+
+- Commit: this work-package commit
+- Results: before the fix the focused capture retained 6 mesh, 6 material, 4 shader, 6 texture, 6 instance and RenderingDevice RIDs (30 resources); after the fix focused and seven-checkpoint captures retained none
+- Remaining risks: D3D12 descriptor fallback remains unrelated; Godot can report a timing-dependent anonymous `RefCounted` whose reference count is already zero, without retained Node/resource/RID ownership
+- Documentation updated: `RENDER.md`, `docs/modules/world-renderer.md`, `docs/M00_FINAL_PAIRED_EVIDENCE.md`; existing renderer shutdown sequence updated, no new diagram required

docs/M00_FINAL_PAIRED_EVIDENCE.md

@@ -0,0 +1,76 @@
+# M00 Final Paired Evidence — 2026-07-13
+
+## Scope and result
+
+The final local M00 audit used Godot 4.6.1, the existing cache state, the manifest-standard 8-second readiness wait, and `2560x1440` visual candidates. Seven private WoW 3.3.5a build 12340 references produced fourteen cold/warm comparisons with no missing pair. Private images and generated reports remain outside Git.
+
+Every pair exceeded the strict regression tolerance. Comparator exit code `1` means a valid report with measured visual gaps; only codes outside `0/1` are infrastructure failures. This evidence establishes coverage and current differences, not renderer parity.
+
+```powershell
+.\tools\run_render_baseline.ps1 `
+  -Output user://render_baseline_m00_complete_2026-07-13 `
+  -CacheState existing `
+  -WaitSeconds 8 `
+  -MeasureSeconds 0.5 `
+  -ReferenceCheckpointDirectory .\sources\OpenWCReferenceCheckpoints `
+  -VisualComparisonReport user://render_baseline_m00_complete_2026-07-13/visual_comparison.json `
+  -VisualComparisonWidth 2560 `
+  -VisualComparisonHeight 1440
+```
+
+The initial shell timeout detached the canonical capture process after it started. The process completed all fourteen canonical captures and wrote `report.json`; the documented visual-capture and comparator stages were then invoked separately with the same arguments. The final output directory contains both complete reports.
+
+## Paired metrics
+
+| Checkpoint | Pass | Mean perceptual error | Changed-pixel ratio | Classification |
+|---|---|---:|---:|---|
+| `elwynn_adt_boundary` | cold | 0.087258 | 0.634580 | Composition plus terrain/vegetation/material gaps |
+| `elwynn_adt_boundary` | warm | 0.087255 | 0.634598 | Stable against revisit |
+| `elwynn_sky_dusk` | cold | 0.094778 | 0.575956 | Approximate paired camera; sky color, lighting and scene-composition gaps |
+| `elwynn_sky_dusk` | warm | 0.094757 | 0.575898 | Stable against revisit |
+| `elwynn_terrain_overview` | cold | 0.174137 | 0.851791 | Largest composition/framing gap |
+| `elwynn_terrain_overview` | warm | 0.176955 | 0.861271 | Stable gap, not a cold-load artifact |
+| `elwynn_waterfall_liquid` | cold | 0.088629 | 0.706522 | Framing, terrain, liquid and lighting gaps |
+| `elwynn_waterfall_liquid` | warm | 0.087398 | 0.693690 | Stable against revisit |
+| `goldshire_dense_m2` | cold | 0.066893 | 0.575257 | Smallest available paired gap; vegetation/material differences remain |
+| `goldshire_dense_m2` | warm | 0.066927 | 0.575592 | Stable against revisit |
+| `goldshire_inn_large_wmo` | cold | 0.101224 | 0.701835 | WMO is ready; manual framing and presentation gaps remain |
+| `goldshire_inn_large_wmo` | warm | 0.101206 | 0.701855 | Stable against revisit |
+| `gryphon_roost_native_animation` | cold | 0.119467 | 0.882073 | Same build-12340 asset, different real/synthetic placement and uncontrolled animation phase |
+| `gryphon_roost_native_animation` | warm | 0.119471 | 0.882076 | Full-frame metric is diagnostic only; animation observation is human evidence |
+
+Cold and warm values are close for all seven references. The measured gaps are therefore not explained by first-visit cache readiness. Goldshire timing evidence separately demonstrated that the 8-second wait is necessary for WMO publication.
+
+## Animation and sky provenance
+
+The private source corpus preserves exact screenshot bytes; no resize, crop or generated content was used:
+
+| Evidence | SHA-256 | Interpretation |
+|---|---|---|
+| `elwynn_sky_dusk.jpg` | `1fb06836cacae714c76eaf9b8bf7c292fd5e2f2df33eef821bfe4c4a6bd97072` | Build 12340, Elwynn camera position, server time approximately 19:00, UI hidden |
+| `gryphon_roost_native_animation.jpg` | `d3488c741cb8dfe7212198e48332842631a951862a4388b67045338f37357447` | Build 12340 real-world `GryphonRoost01` placement, first visible wing phase |
+| diagnostic animation phase B | `1025572c7ab9fa0d2e500ff026be46eb9c1e6e5447116dbdc3675dd5f44574ab` | Same camera three seconds later; visibly raised wing phase |
+| diagnostic placement context | `b12c0b713a7d75123ed30f5a316b33537331442cd56615aea69214c7cc4e7a42` | Two real Lakeshire placements and surrounding build-12340 scene |
+
+An ADT scan found no `GryphonRoost01` placement in Goldshire. The M00 checkpoint deliberately injects the model there as a synthetic probe. The original-client observation therefore uses two real Lakeshire placements. The three-second pose change establishes `Observed` native animation behavior, but the full-frame comparison cannot establish spatial or animation-phase parity.
+
+## Fidelity interpretation
+
+M00 now has private build 12340 evidence for every required coverage category: terrain, ADT boundary, dense M2, large WMO, liquid, animated M2 and sky transition. The sky reference is approximately paired by map position/time and manually approved composition. Animated-M2 evidence is asset-paired but placement-unpaired because the OpenWC probe is synthetic.
+
+The reports demonstrate current differences and do not claim `1:1`. Camera calibration experiments also show that unknown manual yaw/pitch/FOV cannot be recovered reliably from full-frame color error alone. Region scoring reduces terrain bias but still requires semantic landmark approval.
+
+## Operational risks
+
+A follow-up shutdown ownership fix removed the GUI renderer leaks found by the final visual capture. Before the fix, the focused animated-M2 checkpoint retained 6 mesh, 6 material, 4 shader, 6 texture, 6 instance and associated RenderingDevice RIDs, for 30 resources in use. After the fix, both the focused 8-second capture and a seven-checkpoint single-pass capture completed without Node, resource or RID leak diagnostics.
+
+Godot can still emit a timing-dependent ObjectDB warning for one or more anonymous `RefCounted` objects whose reported reference count is already zero. The warning has no retained Node, renderer resource or RID ownership in the verified captures. It remains recorded as an engine-teardown diagnostic rather than being hidden, but it is distinct from the resolved GUI renderer leak.
+
+## Integrator decision checklist
+
+- Seven required reference checkpoints: cold/warm report complete, `compared=14`, `missing=0`.
+- Animated M2: build-12340 native motion observed in two phases; synthetic/real placement mismatch explicitly classified.
+- Sky transition: build-12340 19:00 evidence present with approximate paired camera and human-approved composition.
+- Renderer visual differences: registered as gaps, not parity claims.
+- GUI renderer RID/resource diagnostics: resolved by explicit shutdown ownership and covered by a headless regression verifier; the residual zero-reference `RefCounted` engine-teardown warning remains tracked separately.
+- Target status, target Evidence and `OPENWC_TARGET_DONE` marker: integrator-only actions.

docs/modules/world-renderer.md

@@ -143,7 +143,8 @@ sequenceDiagram
     Stream->>Render: evict outside retention range
     Stream->>Worker: shutdown: wait for WorkerThreadPool tasks
     Stream->>Stream: shutdown: finish registered ResourceLoader requests
-    Stream->>Render: clear queues, nodes, caches and RIDs
+    Stream->>Render: clear queues and owned tree nodes/RIDs
+    Stream->>Stream: free detached prototypes and release resource caches
 ```
 
 ## Ownership, threading and resources
@@ -153,6 +154,8 @@ sequenceDiagram
 - Parsed/grouped results передаются обратно через guarded result queues.
 - Mesh/material/node/RID finalization выполняется main thread и ограничивается exported budgets.
 - Shutdown/map switch обязан дождаться WorkerThreadPool jobs и зарегистрированных threaded ResourceLoader requests до очистки очередей, nodes, caches и RIDs.
+- Parented streamed nodes остаются owned by SceneTree during shutdown; detached M2/WMO prototype caches are explicitly `free()`d after asynchronous work drains.
+- A temporary liquid root with no generated child geometry is explicitly freed by `ADTBuilder` instead of being returned or abandoned.
 - Cache resources считаются immutable после публикации.
 
 ## Errors, cancellation and recovery
@@ -165,7 +168,7 @@ sequenceDiagram
 | Main-thread hitch | Named section timing | Frame spike, work remains queued | `HITCH` log | Lower budget/fix finalize path |
 | D3D12 descriptor exhaustion | Rendering backend error | Render failure/fallback backend | Godot error + baseline report | Dedup resources/fix settings |
 | Teleport/map change | Focus/session transition | Old jobs become stale | Target/session generation | Cancel/drop stale results |
-| Shutdown leak | Godot leak/RID diagnostics | Resource retained | Verbose shutdown report | Drain owned worker and resource requests before clearing registries |
+| Shutdown leak | Godot leak/RID diagnostics | Resource retained | Verbose shutdown report + cache shutdown verifier | Drain requests, preserve tree ownership, then free detached prototypes and clear resource caches |
 
 ## Configuration and capabilities
 
@@ -197,7 +200,7 @@ Exact exported settings and cache versions remain documented in [`../../RENDER.m
 
 - Unit/contract tests: material mapping, unique-ID dedupe, placement probes, baseline manifest, five-point coordinate calibration, synthetic perceptual checkpoint diff, camera-pose grid plan.
 - Integration/E2E: Eastern Kingdoms/Kalimdor streaming scenes and seven cold/warm checkpoints.
-- Fidelity evidence: пять локальных build 12340 reference JPG откалибровали terrain/ADT/M2/WMO/liquid viewpoints; automated paired-image metrics exist, но synthetic animation/dusk и полный human approval ещё не закрыты.
+- Fidelity evidence: семь локальных build 12340 reference JPG покрывают terrain/ADT/M2/WMO/liquid/sky и реальный `GryphonRoost01`; automated cold/warm metrics имеют `compared=14`, `missing=0`. Sky использует приблизительно paired camera, а animated-M2 остаётся asset-paired/placement-unpaired из-за synthetic probe.
 - Performance budgets: M00 report records cold/warm p95 and max hitch; no final acceptance threshold yet.
 - Manual diagnostics: [`../RENDER_BASELINE.md`](../RENDER_BASELINE.md) and [`../../RENDER.md`](../../RENDER.md).
 
@@ -234,7 +237,8 @@ Exact exported settings and cache versions remain documented in [`../../RENDER.m
 - A ready-scene Goldshire 3x3 grid improved mean error from `0.101402` at zero offsets to `0.087952` at yaw `10`/pitch `-10`; pitch/FOV refinement remains required before manifest calibration.
 - Joint Goldshire pitch/FOV refinement had no human-approved optimum: full-frame error preferred grass-heavy FOV 62/pitch -25, while FOV 38/pitch -10 better matched building scale. Landmark/region scoring is required before calibration.
 - Common-region scoring `[300,100,1400,650]` moved Goldshire's numeric optimum to FOV 38/yaw 10/pitch -15, but human inspection still found an oversized/cropped facade; semantic landmarks or masks remain required.
-- D3D12 descriptor issues remain; the capture-path anonymous `RefCounted` shutdown leak is regression-covered by a clean verbose dry-run, while other RID/resource diagnostics still require independent evidence.
+- Final paired audit compared fourteen cold/warm images for seven private build 12340 references with zero missing pairs; mean error ranges `0.066893..0.176955`. Animated-M2 native motion is observed across two original-client phases, but its real Lakeshire placement differs from the synthetic Goldshire probe. Sky-transition evidence uses the checkpoint map position/time with manually approved approximate framing.
+- D3D12 descriptor issues remain. Explicit shutdown ownership removes the GUI capture's Node/resource/RID leaks in focused and seven-checkpoint runs; a timing-dependent ObjectDB warning for anonymous zero-reference `RefCounted` objects remains as a separate engine-teardown diagnostic.
 - M2/WMO/material/particle/ribbon/portal parity incomplete.
 - Public API is mostly exported configuration rather than stable contracts.
 
@@ -252,6 +256,7 @@ Exact exported settings and cache versions remain documented in [`../../RENDER.m
 | `src/tools/build_*cache.gd`, `src/tools/bake_*cache.gd` | Offline cache generation |
 | `tools/run_render_baseline.ps1` | Unified M00 baseline runner |
 | `src/tools/compare_render_checkpoints.gd` | Offline JPG/PNG paired-image perceptual metrics and JSON pass/fail report |
+| `src/tools/verify_render_runtime_cache_shutdown.gd` | Headless ownership regression for detached runtime prototypes, resource caches and empty liquid roots |
 | `src/tools/capture_render_checkpoints.gd` | Deterministic no-roll checkpoint camera, performance and visual capture |
 | `tools/sweep_render_checkpoint_camera_pose.ps1` | Offline yaw/pitch capture grid and paired-error ranking |
 | `src/tools/verify_render_coordinate_calibration.gd` | Build 12340 camera-coordinate golden point round-trip diagnostic |
@@ -263,5 +268,6 @@ Exact exported settings and cache versions remain documented in [`../../RENDER.m
 - [`../../RENDER.md`](../../RENDER.md)
 - [`../RENDER_BASELINE.md`](../RENDER_BASELINE.md)
 - [`../CAMERA_POSE_SWEEP.md`](../CAMERA_POSE_SWEEP.md)
+- [`../M00_FINAL_PAIRED_EVIDENCE.md`](../M00_FINAL_PAIRED_EVIDENCE.md)
 - [`../../targets/roadmap/02-rendering-and-graphics.md`](../../targets/roadmap/02-rendering-and-graphics.md)
 - [`../GODOT_BEST_PRACTICES.md`](../GODOT_BEST_PRACTICES.md)

src/scenes/streaming/streaming_world_loader.gd

@@ -374,6 +374,39 @@ func _exit_tree() -> void:
 	_shutting_down = true
 	_wait_for_tile_tasks()
 	_clear_streamed_world()
+	_release_runtime_caches_for_shutdown()
+
+
+## Frees detached prototype nodes and releases render resources owned by this
+## loader after all asynchronous work has completed. Map refreshes retain these
+## caches; only scene shutdown destroys them.
+func _release_runtime_caches_for_shutdown() -> void:
+	_free_detached_node_cache(_m2_scene_cache)
+	_free_detached_node_cache(_m2_animated_scene_cache)
+	_free_detached_node_cache(_wmo_prototype_cache)
+	_m2_mesh_cache.clear()
+	_m2_static_animation_cache.clear()
+	_m2_missing_cache.clear()
+	_wmo_render_cache.clear()
+	_wmo_scene_resource_cache.clear()
+	_wmo_render_missing_cache.clear()
+	_wmo_scene_cache_missing.clear()
+	_wmo_missing_cache.clear()
+	_shared_tex_cache.clear()
+
+
+func _free_detached_node_cache(node_cache: Dictionary) -> void:
+	for cached_value in node_cache.values():
+		if not (cached_value is Node):
+			continue
+		var cached_node := cached_value as Node
+		if not is_instance_valid(cached_node):
+			continue
+		if cached_node.is_inside_tree():
+			cached_node.queue_free()
+		else:
+			cached_node.free()
+	node_cache.clear()
 
 
 func _apply_quality_preset() -> void:
@@ -2781,15 +2814,27 @@ func _clear_streamed_world() -> void:
 				_free_render_instance(chunk_rid)
 		var tile_root: Node = state.get("root", null)
 		if tile_root:
-			tile_root.queue_free()
+			_queue_free_streamed_node(tile_root)
 
 	_tile_states.clear()
 	_clear_tile_mesh_cache()
 	_clear_quality_terrain_mesh_cache()
 	_clear_wmo_registry()
-	_ensure_world_wmo_root()
-	if enable_wmo:
-		_place_global_wmo()
+	if not _shutting_down:
+		_ensure_world_wmo_root()
+		if enable_wmo:
+			_place_global_wmo()
+
+
+## During world shutdown, parented nodes are recursively owned by the world
+## root. Queueing them separately from `_exit_tree()` can detach them from that
+## deletion and leave empty Node objects alive at engine teardown.
+func _queue_free_streamed_node(node: Node) -> void:
+	if node == null or not is_instance_valid(node):
+		return
+	if _shutting_down and node.is_inside_tree():
+		return
+	node.queue_free()
 
 
 func _reset_terrain_root_children() -> void:
@@ -3889,10 +3934,10 @@ func _clear_wmo_registry() -> void:
 		var entry: Dictionary = entry_variant
 		var node: Node = entry.get("node", null)
 		if node and is_instance_valid(node):
-			node.queue_free()
+			_queue_free_streamed_node(node)
 	_wmo_registry.clear()
 	if _world_wmo_root and is_instance_valid(_world_wmo_root):
-		_world_wmo_root.queue_free()
+		_queue_free_streamed_node(_world_wmo_root)
 	_world_wmo_root = null
 
 
@@ -4458,7 +4503,7 @@ func _cancel_m2_build_job(key: String) -> void:
 	var job: Dictionary = _m2_build_jobs[key]
 	var root: Node = job.get("root", null)
 	if root != null and is_instance_valid(root):
-		root.queue_free()
+		_queue_free_streamed_node(root)
 	_m2_build_jobs.erase(key)
 
 

src/tools/verify_render_runtime_cache_shutdown.gd

@@ -0,0 +1,77 @@
+extends SceneTree
+
+const STREAMING_WORLD_LOADER := preload("res://src/scenes/streaming/streaming_world_loader.gd")
+const ADT_BUILDER := preload("res://addons/mpq_extractor/loaders/adt_builder.gd")
+
+var _failures := 0
+
+
+func _initialize() -> void:
+	_verify_empty_water_root_ownership()
+	_verify_runtime_cache_ownership()
+
+	if _failures > 0:
+		push_error("Render runtime cache shutdown verification failed: %d issue(s)" % _failures)
+		quit(1)
+		return
+	print("Render runtime cache shutdown verification passed.")
+	quit(0)
+
+
+func _verify_empty_water_root_ownership() -> void:
+	var node_count_before := int(Performance.get_monitor(Performance.OBJECT_NODE_COUNT))
+	var empty_water_root: Node3D = ADT_BUILDER.build_tile_water_scene({"chunks": []}, Vector3.ZERO)
+	var node_count_after := int(Performance.get_monitor(Performance.OBJECT_NODE_COUNT))
+	_expect(empty_water_root == null, "empty ADT liquid payload should not produce a water root")
+	_expect(node_count_after == node_count_before, "empty ADT liquid payload should release its temporary water root")
+
+
+func _verify_runtime_cache_ownership() -> void:
+	var loader: Node = STREAMING_WORLD_LOADER.new()
+	var static_m2_prototype := Node3D.new()
+	var animated_m2_prototype := Node3D.new()
+	var wmo_prototype := Node3D.new()
+
+	loader.set("_m2_scene_cache", {"static": static_m2_prototype})
+	loader.set("_m2_animated_scene_cache", {"animated": animated_m2_prototype})
+	loader.set("_wmo_prototype_cache", {"wmo": wmo_prototype})
+	loader.set("_m2_mesh_cache", {"mesh": ArrayMesh.new()})
+	loader.set("_m2_static_animation_cache", {"static": true})
+	loader.set("_m2_missing_cache", {"missing": true})
+	loader.set("_wmo_render_cache", {"render": Resource.new()})
+	loader.set("_wmo_scene_resource_cache", {"scene": PackedScene.new()})
+	loader.set("_wmo_render_missing_cache", {"missing": true})
+	loader.set("_wmo_scene_cache_missing", {"missing": true})
+	loader.set("_wmo_missing_cache", {"missing": true})
+	loader.set("_shared_tex_cache", {"texture": ImageTexture.new()})
+
+	loader.call("_release_runtime_caches_for_shutdown")
+
+	_expect(not is_instance_valid(static_m2_prototype), "static M2 prototype should be freed")
+	_expect(not is_instance_valid(animated_m2_prototype), "animated M2 prototype should be freed")
+	_expect(not is_instance_valid(wmo_prototype), "WMO prototype should be freed")
+	for cache_name in [
+		"_m2_scene_cache",
+		"_m2_animated_scene_cache",
+		"_wmo_prototype_cache",
+		"_m2_mesh_cache",
+		"_m2_static_animation_cache",
+		"_m2_missing_cache",
+		"_wmo_render_cache",
+		"_wmo_scene_resource_cache",
+		"_wmo_render_missing_cache",
+		"_wmo_scene_cache_missing",
+		"_wmo_missing_cache",
+		"_shared_tex_cache",
+	]:
+		var cache: Dictionary = loader.get(cache_name)
+		_expect(cache.is_empty(), "%s should be empty after shutdown" % cache_name)
+
+	loader.free()
+
+
+func _expect(condition: bool, message: String) -> void:
+	if condition:
+		return
+	_failures += 1
+	push_error(message)

tools/run_render_baseline.ps1

@@ -40,11 +40,21 @@ function Invoke-GodotStep {
     }
 }
 
+function Invoke-GodotComparisonStep {
+    param([string]$Name, [string[]]$Arguments)
+    Write-Host "[$Name] $GodotPath $($Arguments -join ' ')"
+    $process = Start-Process -FilePath $GodotPath -ArgumentList $Arguments -NoNewWindow -Wait -PassThru
+    if ($process.ExitCode -notin @(0, 1)) {
+        throw "$Name failed with exit code $($process.ExitCode)"
+    }
+}
+
 Push-Location $repoRoot
 try {
     Invoke-GodotStep "project-load" @("--headless", "--path", ".", "--quit")
     Invoke-GodotStep "render-materials" @("--headless", "--path", ".", "--script", "res://src/tools/verify_render_materials.gd")
     Invoke-GodotStep "m2-unique-dedupe" @("--headless", "--path", ".", "--script", "res://src/tools/verify_m2_unique_dedupe.gd")
+    Invoke-GodotStep "runtime-cache-shutdown" @("--headless", "--path", ".", "--script", "res://src/tools/verify_render_runtime_cache_shutdown.gd")
     Invoke-GodotStep "baseline-manifest" @("--headless", "--path", ".", "--script", "res://src/tools/verify_render_baseline_manifest.gd")
     Invoke-GodotStep "coordinate-calibration" @("--headless", "--path", ".", "--script", "res://src/tools/verify_render_coordinate_calibration.gd")
 
@@ -80,7 +90,7 @@ try {
 			"--viewport-height", $VisualComparisonHeight.ToString()
 		)
 		Invoke-GodotStep "visual-capture" $visualCaptureArgs
-        Invoke-GodotStep "visual-comparison" @(
+        Invoke-GodotComparisonStep "visual-comparison" @(
             "--headless", "--path", ".",
             "--script", "res://src/tools/compare_render_checkpoints.gd",
             "--",