fix(lab): seal RAV004 spatial replay transport

docs/handoff/2026-08-30_RAV004_CANONICAL_LAB_REPLAY_AUDIT.md

@@ -22,6 +22,29 @@ Window structure, switching, seek, buffering and spatial scene code are shared.
 
 ## Why the previous LAB failed
 
+### RAV004 did not use the accepted replay data plane
+
+The working M4/Hologravity LAB uses sealed binary numeric tracks, retained scene
+state and bounded JSON metadata. RAV004 reused the visual component but silently
+left its custom timeline endpoint on the JSON fallback. Each eight-frame spatial
+window therefore transferred approximately 1.2-2.9 MB and took 0.87-1.37 s to
+produce while representing only about 0.84 s of playback. The next request
+aborted and replaced the previous one before spatial state could catch up.
+
+The user screenshot captured the failure exactly: camera frame 366 was active
+while the last delivered spatial evidence was frame 230, a 136-frame gap. The
+backend also reopened and indexed the 6830-entry semantic ZIP for every requested
+mask and proposal frame. This explains why the same canonical viewer was smooth
+for Hologravity but stalled for RAV004: the window and interaction code were
+shared, but the data-plane contract was not.
+
+RAV004 now publishes the same `missioncore.recorded-spatial-playback/v1`
+contract as the accepted LAB: an immutable Float32 map-point track, camera-frame
+offsets and 24-frame binary chunks. JSON chunks contain bounded frame metadata
+only; retained Local SLAM is reconstructed from exact source increments in the
+shared client. The semantic ZIP handle and member index are cached per immutable
+artifact instead of being reparsed per frame.
+
 ### Video and spatial state had different clocks
 
 The removed RAV004 viewer advanced an animation/host clock even when the browser
@@ -77,8 +100,11 @@ dozens of seconds old. The buffer now loads the active chunk first, the precedin
 chunk second and the next chunk as prefetch. The scene selects the latest proven
 source increment whose sequence is not later than the active camera frame.
 
-At the final UI check, camera frame 189 causally held spatial frame 184. Before
-the fix the same point could hold frame 16.
+At the final UI check, playback restarted at frame 1, then ran continuously past
+frame 462. At frame 188 the DDRNet mask was frame 188, proposals were frame 187
+and spatial state was delivered without buffering; the one-frame proposal delay
+is the recorded causal overlay, not stale UI state. Before the transport fix the
+user's run had already fallen 136 frames behind by camera frame 366.
 
 ## Capability ledger
 
@@ -104,12 +130,17 @@ full-TGS result is also not reused because it has a different source identity an
 Measured on the canonical local service and current immutable artifacts:
 
 - replay launch POST: 3.55 s on first opening;
-- timeline metadata: 0.02 s warm;
-- active spatial chunk, eight camera frames: 35.17 s first process-local RRD
-  index build, 0.67 s warm, approximately 3.81 MB;
-- UI replay: passed the previously deterministic 11.422 s decoder stop, then
-  continued to 59 s with media and timeline advancing together;
-- operator reset seek: 16.4 s to 0 s, one mounted media worker, successful;
+- first binary playback-manifest build after a service restart: 32.33 s while
+  the process-local RRD point track is materialized; warm manifest: 0.18-0.20 s;
+- full retained point track: 3,893,445 Float32 map points, 46,721,340 bytes,
+  divided into 285 immutable 24-frame chunks;
+- representative active binary chunks: 155-224 KB at 9-10 ms;
+- representative 24-frame metadata chunks: 24-32 KB at 0.78-1.0 s, covering
+  about 2.4 s of playback;
+- cached semantic-mask reads: 5.6-7.8 ms instead of approximately 80 ms;
+- UI replay: reset to frame 1 and ran continuously beyond frame 462 with camera,
+  semantic overlay and retained spatial state advancing together;
+- operator reset seek: successful, one mounted media worker;
 - browser console after the acceptance run: no warnings or errors.
 
 The first RRD index is still process-local rather than a persistent disk cache.
@@ -147,13 +178,14 @@ deploy it.
 
 ## Acceptance performed
 
-- 44 focused backend tests passed;
-- 37 frontend replay, buffering and LAB contract tests passed;
+- 12 focused backend spatial/API tests passed;
+- 16 focused frontend replay transport/manifest tests passed;
 - TypeScript project typecheck passed;
 - production Vite build passed (only existing large-chunk warnings);
 - `git diff --check` passed;
-- live browser run verified the shared controls, disabled unsealed TGS/3D
-  semantics, continuous media recovery, causal spatial hold and clean console.
+- live browser run verified reset seek, the shared controls, disabled unsealed
+  TGS/3D semantics, continuous playback through the former failing interval,
+  causal spatial hold and a clean console.
 
 Visual QA: `docs/handoff/2026-08-30_RAV004_CANONICAL_LAB_QA.jpg`.