fix(diagnostics): keep UI scripts on game-loop time

Preserve sub-millisecond frame deltas with a command-local compensated clock so uncapped diagnostic routes do not outrun materialization and settling. Keep sleep boundaries inclusive, wait timeouts strict, and handle long-duration and invalid-delta cases deterministically.

Record the aligned seven-destination connected route with a full turn at every stop. Release build succeeds and all 6,308 tests pass with five intentional skips.

Co-authored-by: OpenAI Codex <codex@openai.com>

docs/ISSUES.md

@@ -126,18 +126,51 @@ stable radix, avoiding a second immediate-buffer pack for deferred transparent
 instances, packing per-instance light sets, caching static picking descriptors,
 and using checked direct group handles. Historical material groups now retire
 after one complete unused frame (not after one `Draw`, because landscape slices
-and paperdoll share the dispatcher) and release their list capacity. That dense
-sample reached roughly 5.3–6.2 ms CPU / 7.1–8.3 ms p95 and 3.2–3.6 ms GPU before
-the final group-retirement review correction.
+and paperdoll share the dispatcher) and release their list capacity. Before the
+final group-retirement review correction, that fixed dense sample reached
+5.3–6.2 ms CPU p50, 7.1–8.3 ms CPU p95, and 3.2–3.6 ms GPU p50.
 
 The earlier software limiter also used coarse `Thread.Sleep`, whose nominal
 one-millisecond waits measured about 15.98 ms on this host. A reusable Windows
 high-resolution waitable timer now paces to the monitor deadline and is reported
 as `pace` by the frame profiler. Scripted UI automation no longer enables the
 allocation-heavy GPU-stream dump implicitly, and graceful shutdown now wakes
-all persistent mesh workers without resetting their shared wake signal. The
-post-review connected stress rerun remains pending because local ACE was not
-listening; sparse-scene 239 FPS samples are not substituted for that gate.
+all persistent mesh workers without resetting their shared wake signal. The UI
+automation clock now uses command-local compensated accumulation, retaining
+sub-millisecond frame deltas instead of rounding every uncapped frame up to one
+millisecond. Exact sleep and strict timeout boundaries therefore remain stable
+across high-refresh frames and long scripts.
+
+The post-review connected rerun passed on 2026-07-19. It used the complete
+Caul → Sawato → Rynthid → Aerlinthe → Sawato → Holtburg → Caul route, waited for
+each actual materialization, allowed at least five seconds of destination
+streaming, synchronized to the next profiler boundary, and then held the retail
+turn-right action through one dedicated interval (4.93–5.07 seconds, more than
+one full visibility sweep) at every stop. Each
+turn window and its following stationary window were captured on separate
+profiler boundaries. The uncapped Release client completed all seven portals
+and seven turns and shut down gracefully, with no exception, WER report, or AMD
+driver reset observed in the process logs, Windows event logs, or crash-dump
+directory. Returned dense Caul contained about 21,024 visible entities and
+measured 3.8 ms median / 5.0 ms p95 CPU and 2.6 ms median / 3.2 ms p95 GPU while
+stationary (241 FPS in the title sample). Its complete turning window measured
+3.0 ms median / 4.9 ms p95 CPU and 2.0 ms median / 3.4 ms p95 GPU.
+
+Across all stationary destinations, CPU p50/p95 ranged from 1.1–4.0 /
+1.6–5.1 ms and GPU p50/p95 from 0.7–2.8 / 1.0–3.4 ms. Across the turn
+windows, those ranges were 1.0–3.1 / 1.8–4.9 ms CPU and
+0.7–2.0 / 1.2–3.4 ms GPU. Working/private memory was 939.8/1,438.3 MiB on the
+first Caul dwell; the highest stationary sample was 1,065.0/1,660.6 MiB in
+Holtburg, followed by 1,034.8/1,630.5 MiB on the Caul return. The samples were
+non-monotonic, but one circuit is not claimed as a new lifetime plateau proof.
+Startup reached a transient 240.2 ms maximum, and the largest later
+destination-load window reached 81.9 ms CPU; all isolated turn and stationary
+windows recovered to low-millisecond frame times, so this run is not described
+as stall-free.
+
+The run was performed over RDP, but software pacing was explicitly disabled and
+`pace` remained zero, so these are engine-throughput measurements rather than
+remote-display refresh measurements. The lifestone/particle visual gate remains.
 
 **Files:** `src/AcDream.App/Rendering/RetailAlphaQueue.cs`;
 `src/AcDream.App/Rendering/Wb/WbDrawDispatcher.cs`;
@@ -153,7 +186,13 @@ listening; sparse-scene 239 FPS samples are not substituted for that gate.
 `src/AcDream.Content/BoundedDatObjectCache.cs`;
 `src/AcDream.Content/DecodedTextureCache.cs`;
 `src/AcDream.App/Rendering/FramePacingController.cs`;
-`src/AcDream.App/Rendering/WindowsHighResolutionFramePacingWaiter.cs`.
+`src/AcDream.App/Rendering/WindowsHighResolutionFramePacingWaiter.cs`;
+`src/AcDream.App/Diagnostics/FrameProfiler.cs`;
+`src/AcDream.App/Rendering/GameWindow.cs`;
+`src/AcDream.App/Rendering/Wb/ObjectMeshManager.cs`;
+`src/AcDream.App/Rendering/Wb/CachedBatch.cs`;
+`src/AcDream.App/Rendering/Wb/EntityClassificationCache.cs`;
+`src/AcDream.App/UI/Testing/RetailUiAutomationScriptRunner.cs`.
 
 **Research:**
 `docs/research/2026-07-18-retail-shared-alpha-list-pseudocode.md`;

docs/plans/2026-04-11-roadmap.md

@@ -562,7 +562,7 @@ behavior. Estimated 17–26 days focused work, 3–5 weeks calendar.
 - **Missile/portal VFX campaign Step 9 automated hardening implemented and independently reviewed 2026-07-14; final two-client visual gate pending.** A deterministic 96-owner App fixture drives canonical missiles and effect owners through projectile updates, repeated DAT-effect scheduling, loaded↔pending↔loaded landblock churn, light/particle withdrawal and recovery, accepted deletes, same-GUID generation reuse, never-created F754 queues, and session reset. It asserts balanced logical render registration and zero residual records/bodies/projectiles, spatial buckets/rescues, shadows (including suspended registrations), effect profiles/packets, PhysicsScript FIFOs/anchors/delayed calls, particle bindings/logical IDs/render-pass owners, poses, light-controller/sink/manager owners, and stale record component references. A companion twelve-cycle gate drives exact recall motion `0x10000153` through AnimationSequencer/CallPES, Hidden, deferred remote placement, hydration, and UnHide; a second 96-owner `EntitySpawnAdapter` gate balances actual mesh-adapter reference counts without GL. The pass fixed undrained persistent rescue retention, delayed stale visibility edges, GUID-scoped teardown, create resurrection after a nested delete/reset, non-atomic resource registration, double teardown from a re-entrant session-clear callback, observer failure stranding later visibility edges, and stale outer projection transactions overwriting callback-created replacements. Teardown now removes exact projection references and generation/local-ID owners, uses per-GUID/session lifetime epochs plus per-record projection tokens, drains visibility fan-out before aggregating failures, and finishes or supersedes canonical commits before surfacing observer errors. Core remains GL/backend-free, panels remain on UI abstractions, and projectiles still draw through ordinary `WbDrawDispatcher` live-entity submission rather than a global projectile pass. AP-69 and TS-49 remain; AP-83/AP-91 now explicitly describe only a future mover that enables PerfectClip.
 
 - **Portal/resource lifetime correction 2026-07-18; connected stress gate passed, final visual gate pending.** `LiveEntityLivenessController` ports retail's 25-second leave-visibility destruction queue over canonical live records, cancels expiry for spatially resident or owned objects, and uses holtburger's conservative 384-unit envelope where ACE supplies no client-visible PVS leave event. The broader integration balances generation-scoped appearance/live/landblock/effect/UI/portal owners, bounds DAT/decoded/standalone/composite/atlas/mesh residency, performs incremental mesh-arena migration, and defers GL deletion or slot reuse through three frames of GPU fences. Render/UI scratch storage is reused with hysteretic trimming. No visual range or quality setting was reduced. The final seven-destination connected route cut peak working/private memory from 2,954.5/3,502.3 MiB to 1,493.4/1,969.5 MiB, returned Caul to 1,030.6/1,638.2 MiB, held 125–153 FPS locally through the final dwell, and produced no WER/driver reset. AP-69 is narrowed rather than retired because exact ObjCell PVS and explicit preview lifetimes remain.
-- **Render-thread CPU and high-refresh pacing correction 2026-07-19; automated/review gates passed, connected rerun pending.** Software pacing now uses one reusable Windows high-resolution waitable timer instead of scheduler-quantized `Thread.Sleep`; the permanent profiler attributes that wait as `pace`. Wb keeps exact stable far-to-near alpha order while using a radix key, avoids immediate SSBO packing for deferred transparent objects, packs each eight-light set, caches immutable selection descriptors, and retires historical material groups only after one complete unused frame across every landscape/dynamics/paperdoll draw scope. A fixed dense-Caul sample improved from roughly 9–12 ms CPU to 5.3–6.2 ms without reducing view distance, particles, textures, streaming radius, or other quality. The prior 141.8 FPS result remains scene-specific; local ACE was offline for the post-review connected rerun.
+- **Render-thread CPU and high-refresh pacing correction 2026-07-19; automated/review/connected gates passed, final translucency visual gate pending.** Software pacing now uses one reusable Windows high-resolution waitable timer instead of scheduler-quantized `Thread.Sleep`; the permanent profiler attributes that wait as `pace`. Wb keeps exact stable far-to-near alpha order while using a radix key, avoids immediate SSBO packing for deferred transparent objects, packs each eight-light set, caches immutable selection descriptors, and retires historical material groups only after one complete unused frame across every landscape/dynamics/paperdoll draw scope. A fixed dense-Caul sample improved from roughly 9–12 ms CPU to 5.3–6.2 ms without reducing view distance, particles, textures, streaming radius, or other quality. The corrected post-review uncapped route waited for every materialization and held a full visibility sweep through one dedicated profiler window at all seven destinations. Returned dense Caul (about 21,024 visible entities) measured 3.0/4.9 ms CPU p50/p95 and 2.0/3.4 ms GPU while turning, then 3.8/5.0 ms CPU and 2.6/3.2 ms GPU stationary (241 FPS title sample). First/return Caul memory was 939.8/1,438.3 and 1,034.8/1,630.5 MiB working/private; the highest stationary sample was 1,065.0/1,660.6 MiB in Holtburg, and the client exited gracefully. With pacing disabled (`pace=0`) and title samples reaching 844 FPS, RDP did not refresh-cap engine throughput. UI automation now uses a command-local compensated clock, preserving sub-millisecond frame deltas and exact long-duration sleep/timeout boundaries.
 
 **Reference docs:** `docs/research/retail-ui/00-master-synthesis.md` + slices 01-06. Every AC-specific behavior has a decompiled FUN_ / DAT_ citation.