research: idle workers at 30 m/s (L122, no production change)

98% of idle-with-holes windows wait on the expanded critical set; region compile runs 0.26 ms but its
gate stays pending 18 ms p50. Broad early admission: near +15.4 pt but p99 13.4 ms (over the VR 11 ms
frame); near-only: +9.9 pt but far holes +1.5 pt. Trace instrumentation kept as a patch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016dhxM43dnWvUu2B5Ph1iG1

Standalone/Research/WorldAssemblerM3/Linux122/REPORT.md

@@ -0,0 +1,18 @@
+# Linux 122 — idle workers at speed
+
+**Decision:** keep the L121 admission policy. Neither tested fast-travel relaxation met every coverage and tick limit.
+
+Five diagnostic runs at 30 m/s had 686 idle-worker windows with near tiles missing. In 673 (98%) the expanded critical set was not ready; 462 (67%) also had Core-region compilation pending, and 441 (64%) declined an exact region-dependent near tile. These counts overlap. Far-parent, near-cap, walkable-ring, pawn-collision, and stale-center holds were all zero. Only 13 windows had pending near work and no unadmitted near tile. The desired center was current in every window.
+
+Region compile requests scheduled immediately (request to task submission p50/p95 **0.008/0.021 ms**). UE task queue delay was **0.122/14.856 ms** p50/p95 (max 54.4 ms); task runtime was **0.485/4.572 ms**, with Core compile wall time **0.259/3.679 ms**. The task runs on UE background workers, not the Game Thread. It was submitted while generation workers were idle in 680/1,158 requests. The compile targets the current desired center; there is no velocity look-ahead. The completion gate remained pending **18.216/21.136 ms** p50/p95 while the Game Thread observed completion.
+
+The UE/headless coverage gap is not a denominator mismatch: both use resident near tiles / desired LOD0–2 tiles, with about 1,346–1,348 desired near tiles in UE versus 1,344 in the saved headless runs. Headless was about 82.6%; UE was 45.0%. The measured idle windows point to UE admission. Apply work is small (L121 near apply and proxy p95 were about 0.20 and 0.24 ms), though worker-finish-to-apply wait reached 120 ms p95 and may add secondary delay.
+
+| Fast-travel design | Near change | Far-hole change | Paired p99 change | Worst tick |
+|---|---:|---:|---:|---:|
+| Admit normal near/far shares before expanded critical readiness | +15.42 pp | −0.28 pp | +0.28 ms | 14.52 ms |
+| Admit near LOD0–2 only until critical readiness | +9.89 pp | +1.48 pp | −0.56 ms | 14.38 ms |
+
+The first design exceeded the p99 bar (worst per-run p99 13.36 ms versus L121's 11.00 ms); the second increased far holes. Both stayed under L121's 14.74 ms worst-tick ceiling. No admission behavior was retained, so arrival behavior remains L121.
+
+UE build passed with `-NoUBA -MaxParallelActions=4`. Exact tile and whole-view hashes matched L104 in **24/24** Clang/GCC W1/W8/W12 runs. Fast tier passed (**269 gates, 5 diagnostics**). Logs are under `Saved/Linux/l122`.