Merge branch 'claude/jovial-chebyshev-d1d9da' — Issue #47 close-detail meshes

Brings GfxObjDegradeResolver and the ACDREAM_RETAIL_CLOSE_DEGRADES wiring
that resolves humanoid body parts to their retail close-detail meshes via
DIDDegrade slot 0. User-confirmed visual fix for the bulky/flat humanoid
body bug.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/ISSUES.md

@@ -111,6 +111,12 @@ the same direction. Add a `LastUMUpdateTime` grace window (e.g.
 
 - `docs/research/2026-05-03-remote-anim-cycle/investigation-prompt.md` —
   full background of the four-agent investigation
+- `docs/research/2026-05-06-locomotion-cycle-transitions/investigation-prompt.md` —
+  expansion to the full 7-transition matrix (Run↔Walk forward + backward,
+  Fast↔Slow strafe L+R, direction-flip cases) with TTD-driven workflow
+- `docs/research/2026-05-06-locomotion-cycle-transitions/findings-static.md` —
+  static-analysis findings + scope of the 2026-05-06 candidate fix
+  (case #1, Run↔Walk forward only)
 - This session's diagnostic logs at `tools/diag-logs/walkrun-A1b-*.log`
   (UM_RAW, FWD_WIRE, SETCYCLE traces) confirming ACE's wire pattern
 
@@ -124,6 +130,34 @@ the same direction. Add a `LastUMUpdateTime` grace window (e.g.
 - No spurious cycle thrashing during turning while running (ObservedOmega
   doesn't trigger velocity-bucket changes).
 
+**Progress 2026-05-06 — Shift-toggle cases (#1, #2, #4, #5) fixed; user-verified:**
+
+Five-commit sequence on this branch (`claude/determined-solomon-d0356d`):
+
+| Commit | Effect |
+|---|---|
+| `8fa04af` | First candidate — added `RemoteMotion.LastUMTime` + `ApplyPlayerLocomotionRefinement` with 500 ms UM grace + forward-direction hysteresis. **Ineffective** because the call site lived in dead code for player remotes. |
+| `863d96b` | Skip transition link in SetCycle for direct cyclic-locomotion → cyclic-locomotion. **Reduces queue accumulation** (qCount climbs slower); not the actual case-#1 fix but architecturally correct. |
+| `bb026b7` | Per-tick `[CURRNODE]` diagnostic — exposed that `_currNode` was correctly tracking SetCycle's intent and so the bug was elsewhere. Read-only. |
+| `2653b30` | **Wire `ApplyServerControlledVelocityCycle` into the L.3 M2 player-remote path.** Found via the diag — the existing call site at `OnLivePositionUpdated` line ~3879 was unreachable for players because the L.3 M2 routing returns at line 3755. New synth-velocity computation + call inserted in the player branch. **User-verified working** for forward Run↔Walk via Shift toggle. |
+| `cc62e1c` | Handle backward (`CurrentSpeedMod < 0` → preserve negative sign) and sidestep (low byte 0x0F / 0x10 → keep motion ID, refine magnitude). Backward regression resolved. |
+| `349ba65` | Use `SidestepAnimSpeed` (1.25) instead of `WalkAnimSpeed` (3.12) when computing sidestep magnitude — fix #4's mapping was 2.5× too small for slow strafe. |
+
+**Wire-level finding refuting the original ISSUES.md root-cause hypothesis: Earlier diagnostic claims that ACE broadcasts UMs on Shift toggle were misread.** A clean test (`launch-39-diag2.log`) holding W and toggling Shift while held shows `[FWD_WIRE]` for retail-driven actor only emitting `Ready ↔ Run` transitions — no Walk wire transitions at all, despite a clear walk-pace ↔ run-pace shift visible in `[VEL_DIAG]`. So retail's outbound DOES go silent on HoldKey-only changes. The earlier launch's many Walk↔Run `[FWD_WIRE]` lines came from W press/release cycles with Shift held continuously — different scenarios.
+
+**Verified working (user, 2026-05-06):**
+
+- Forward Run↔Walk via Shift toggle (case #1)
+- Backward Walk slow↔fast via Shift toggle (case #2) — animation matches direction, no rubber-band
+- Strafe-left / strafe-right slow↔fast via Shift toggle (cases #4 / #5) — cadence visibly changes
+
+**Residual / not yet verified:**
+
+- "Not as fast as retail" — ~500 ms `UmGraceSeconds` window adds latency on top of the UP cadence (5–10 Hz). Could be tuned shorter once cases #3 / #6 / #7 are validated.
+- Direction-flip cases (#3 W↔S, #6 A↔D, #7 W↔A/D) — believed to work via direct UM, not explicitly verified yet.
+
+**New related issue filed: #45** — local-player slow-strafe-walk renders too slow. Same `SidestepAnimSpeed` vs `WalkAnimSpeed` mismatch pattern as fix #5, but on the local-player render path (`UpdatePlayerAnimation`), not the observer side.
+
 ## #42 — [DONE 2026-05-05 · ec59a08] Airborne XY drift on observed player remote jumps (~1 m horizontal offset over arc)
 
 **Status:** DONE
@@ -1224,6 +1258,191 @@ If hypothesis (a) is correct, this issue effectively rolls into **#28** — the
 
 ---
 
+## #47 — [DONE 2026-05-06] Humanoid Setup 0x02000001 renders bulky / lacks shape detail vs retail
+
+**Status:** DONE (commit pending)
+**Closed:** 2026-05-06
+**Severity:** MEDIUM (cosmetic — characters readable but visibly different from retail)
+**Filed:** 2026-05-06
+**Component:** rendering / mesh / character animation
+
+**Resolution:** Root cause was that we drew the base GfxObj id from
+Setup / `AnimPartChange` directly. Retail's `CPhysicsPart::LoadGfxObjArray`
+(`0x0050DCF0`) treats that base id as an **entry point to the
+`DIDDegrade` table**; for close/player rendering it draws
+`Degrades[0].Id`, which is the higher-detail mesh that carries the
+bicep / deltoid / shoulder geometry. ACViewer also has this bug —
+that was the key signal it wasn't acdream-specific.
+
+Concrete swaps the resolver now performs:
+- Aluvian Male upper arm `0x01000055` → `0x01001795` (14/17 → 32/60 verts/polys)
+- Aluvian Male lower arm `0x01000056` → `0x0100178F`
+- Heritage variants: `0x010004BF → 0x010017A8`, `0x010004BD → 0x010017A7`,
+  `0x010004B7 → 0x0100179A`, etc.
+
+Fix landed as `GfxObjDegradeResolver`, gated behind
+`ACDREAM_RETAIL_CLOSE_DEGRADES=1` and scoped to humanoid setups
+(34-part with ≥8 null-sentinel attachment slots). User confirmed
+visually 2026-05-06.
+
+Files: `src/AcDream.Core/Meshing/GfxObjDegradeResolver.cs`,
+`src/AcDream.App/Rendering/GameWindow.cs` (wiring), 5 unit tests in
+`tests/AcDream.Core.Tests/Meshing/GfxObjDegradeResolverTests.cs`.
+Research note: `docs/research/2026-05-06-issue-47-close-degrade-pseudocode.md`.
+
+---
+
+### Original investigation (kept for reference)
+
+**Description:** Every humanoid character using Setup `0x02000001`
+(Aluvian Male) renders in acdream with a "bulky, less-defined" silhouette
+compared to retail's view of the same character. Specifically: shoulders
+look smoother/rounder where retail has pointier shoulder pads; back has
+less contour; arms appear puffier. The effect is identical for player
+characters (`+Acdream`, `+Je`) and for humanoid NPCs using the same
+setup (e.g. Woodsman, Sedor Wystan the Blacksmith, Thelnoth Cort).
+Drudges and other monster setups (e.g. `0x020007DD`) render
+identically to retail, so this is *not* a pipeline-wide bug.
+
+The bug is independent of equipment — `+Je` stripped naked still
+shows the same bulky silhouette.
+
+**Investigation 2026-05-06 (~3 hr session, ruled out many hypotheses):**
+
+What was ruled out:
+
+- **0xF625 ObjDescEvent appearance updates being dropped.** Was a real
+  bug for skin/hair colors; fixed in commit e471527. Does not affect
+  the bulky-shape issue (which persists with the fix in place and
+  with no equipment).
+- **Position-pop on equip toggle.** Caused by re-applying with cached
+  spawn's stale position; fixed in same commit. Doesn't affect shape.
+- **Clothing/armor overlapping the base body** (HiddenParts hypothesis).
+  User stripped naked; bulky shape persists.
+- **ParentIndex hierarchy not walked in `SetupMesh.Flatten`.** Setup
+  `0x02000001` has a real hierarchy (`-1, -1, 1, 2, 3, -1, 5, 6, 7, 0,
+  9, 10, 11, 12, 13, 14, 15, 0, ...`), but implementing parent-walk
+  produced **no visible change** — confirming AC's idle animation
+  frames are already in setup-root coordinates, not parent-local.
+- **Equipment / wielded items.** No equipment on `+Je` and bug persists.
+- **Player-specific data flow.** Humanoid NPCs using same setup
+  (Woodsman) show same bug.
+
+What was confirmed (data captured via `ACDREAM_DUMP_CLOTHING=1`):
+
+- Setup `0x02000001`: `setup.Parts.Count = 34`, `flatten.Count = 34`,
+  `APC = 34..38` depending on equipment.
+- All 34 parts emit triangles successfully (no silent GfxObj load
+  failures). Total ~648-700 tris per character.
+- Idle animation frames place parts at sensible humanoid Z-heights
+  (head Z=1.587, mid-body Z=0.5-1.0, ground Z=0.085).
+- Per-part orientations are nearly all 180° around -Z (W≈0,
+  Z≈-1) — a setup-wide coordinate-flip convention. Drudges have
+  varied per-part orientations.
+- `setup.DefaultScale.Count = 0` for both humans and drudges → all
+  parts use Vector3.One scale.
+
+**Working hypotheses (next session):**
+
+1. **Per-vertex normal style.** AC dat may store per-face normals
+   for human GfxObjs (one normal per polygon, copied to all 3
+   vertices) but smooth normals for monster GfxObjs. acdream uses
+   dat normals directly. Test by computing smooth normals from face
+   adjacency and comparing render. User said "not shaders" but the
+   screenshots clearly show smooth-vs-faceted lighting differences.
+2. **Lighting setup.** Cell ambient may be too low, leaving back-
+   facing surfaces in flat shadow. Compare `uCellAmbient` value
+   against retail's behaviour at the same time-of-day.
+3. **Anti-aliasing.** Retail may use MSAA; acdream window may not.
+   Polygon edges in acdream would be visibly stair-stepped, reading
+   as "more faceted" / blockier.
+4. **Surface flags interpretation.** Specific Surface.Type bits for
+   character textures (skin, fabric) may need handling acdream
+   doesn't yet do (e.g. `SmoothShade` flag, or a mip bias).
+
+**Diagnostic infrastructure landed this session** (env-var-gated, no
+runtime cost when off):
+
+- `ACDREAM_DUMP_CLOTHING=1` extended:
+  - `setup.Parts.Count`, `flatten.Count`, `APC` count on header line
+  - `ParentIndex[]` array dump
+  - `DefaultScale[]` array dump
+  - `IdleFrame.Frames[]` per-part Origin + Orientation (first 17 parts)
+  - `EMIT part=NN gfx=0xXX subMeshes=N tris=N` per part
+  - `TOTAL tris=N meshRefs=N` per entity
+
+**Files (suspect surface area for next investigation):**
+
+- `src/AcDream.Core/Meshing/SetupMesh.cs` — Flatten composition
+- `src/AcDream.Core/Meshing/GfxObjMesh.cs` — polygon emission +
+  vertex normal handling (line 142)
+- `src/AcDream.App/Rendering/Shaders/mesh.frag` — lighting eq
+- `src/AcDream.App/Rendering/Shaders/mesh.vert` — normal transform
+
+**Acceptance:** Side-by-side screenshots of `+Acdream` (or any humanoid
+NPC using `0x02000001`) viewed from the same angle in acdream and
+retail show matching silhouette and shape definition.
+
+---
+
+## #46 — Retail observer of acdream sees blippy / laggy movement
+
+**Status:** OPEN
+**Severity:** MEDIUM (degrades external perception of acdream-driven characters)
+**Filed:** 2026-05-06
+**Component:** net / motion (acdream's outbound path: `PlayerMovementController` → `MoveToState` (0xF61C) / `AutonomousPosition` heartbeat → ACE → retail observer)
+
+**Description:** When viewing acdream's local +Acdream character through a parallel retail acclient.exe, the retail observer sees the character's movement as visibly blippy and laggy — position appears to step in discrete jumps rather than translating smoothly. The local acdream view of the same character looks fine, and acdream observing a retail-driven character (after #39 / #45) also looks fine. The degradation is specifically on the **outbound** side: what acdream sends to ACE for relay to other clients.
+
+**Root cause / status:**
+
+Unverified. The likely culprits, ranked by suspected probability:
+
+1. **AutonomousPosition heartbeat cadence.** `memory/project_retail_motion_outbound.md` notes acdream's fixed 200 ms heartbeat is a probable retail mismatch. Retail's `CommandInterpreter::SendPositionEvent` gates on transient_state (Contact + OnWalkable + valid Position) and may broadcast at a different cadence — fewer / more / variable. If acdream sends too rarely, observer dead-reckons too long between updates and visibly stutters when each AutoPos arrives.
+2. **MoveToState send conditions.** `PlayerMovementController.cs:813-840` decides when a fresh MoveToState fires (state-change detection). If important transitions are missed (e.g., direction changes that don't flip ForwardCommand/SidestepCommand), the observer's last-known motion stays stale and AutoPos updates blip the body to the new authoritative position.
+3. **InstanceSequence / ObjectMovement sequence counters.** ACE rejects out-of-order packets. If acdream's sequence stamping is off, ACE silently drops some packets; observer dead-reckons through the gap.
+4. **Velocity field absent on AutoPos.** ACE relays UPs without HasVelocity for player characters (per `OnLivePositionUpdated` comment). Observer's dead-reckoning between UPs may extrapolate using stale velocity, producing visible position drift that snaps back on the next UP — exactly the blippy pattern.
+
+**Verification approach:**
+
+- Run two retail clients + one acdream client. Drive acdream; observe acdream's character on retail #1 and on retail #2 (both retail observers see the same wire). Compare to a retail-driven character observed from the same retail clients — does it look smooth there? If yes, the issue is acdream-outbound-specific. If both look blippy, it's something on the ACE side (less likely).
+- cdb-attach a retail observer client and breakpoint `MovementManager::unpack_movement` to count UPs and UMs received per second from the acdream-driven character vs from another retail character. The cadence delta will identify which packet stream is misbehaving.
+- Compare acdream's outbound packet timing against holtburger's `client/movement/system.rs` heartbeat logic — that's the closest known-working reference for how a non-retail client should pace its outbound.
+
+**Files:**
+
+- `src/AcDream.App/Input/PlayerMovementController.cs` — outbound state-change detection + heartbeat
+- `src/AcDream.Core.Net/WorldSession.cs` — sequence counters + send path
+- `src/AcDream.Core.Net/Net/Outbound/...MoveToState.cs` / `AutonomousPosition.cs` — wire builders
+- `references/holtburger/crates/holtburger-core/src/client/movement/system.rs` — reference cadence
+
+**Acceptance:**
+
+- Side-by-side comparison: retail observer of acdream-driven character and retail observer of retail-driven character look equally smooth during running, walking, sidestepping, turning, and stopping.
+- No visible "step" pattern when acdream-driven character translates between AutoPos updates.
+
+**Cross-reference:**
+
+- `memory/project_retail_motion_outbound.md` — 2026-05-01 cdb live trace of retail's outbound (`CommandInterpreter::SendMovementEvent` for WASD, `Event_Jump` per-frame while charging).
+- CLAUDE.md "Outbound motion wire format" — the `WalkForward + HoldKey.Run` ↔ `RunForward` auto-upgrade ACE applies on broadcast.
+
+---
+
+## #45 — [DONE 2026-05-06 · e9e080d] Local +Acdream sidestep walking renders too slow
+
+**Status:** DONE
+**Closed:** 2026-05-06
+**Commit:** `e9e080d`
+**Component:** physics / animation (local player path: `UpdatePlayerAnimation`)
+
+**Resolution:** `PlayerMovementController.cs:871` computes `localAnimSpeed` as raw `runRate || 1.0`, but ACE's `BroadcastMovement` converts the inbound `MoveToState.SidestepSpeed` via `speed × 3.12 / 1.25 × 0.5` (`Network/Motion/MovementData.cs:124-131`). Observer-side cycles play at the ACE-scaled value (~1.248 slow / ~3.0 fast clamped); the local cycle was playing at the raw 1.0 / runRate — about 80% of retail cadence for slow strafe.
+
+`UpdatePlayerAnimation` now multiplies `animSpeed` by `WalkAnimSpeed / SidestepAnimSpeed × 0.5 = 1.248` when `animCommand` is `SideStepLeft / Right` (low byte 0x0F or 0x10). User-verified: local strafe cadence matches retail / observer-side rendering.
+
+**Original investigation note (preserved):** Same constant mismatch pattern as #39 fix #5 (commit `349ba65`) but on the local-player render path instead of the observer-side `ApplyPlayerLocomotionRefinement` — both fixed by aligning the speedMod base to ACE's wire formula.
+
+---
+
 ---
 
 # Recently closed

docs/research/2026-05-06-issue-47-close-degrade-pseudocode.md

@@ -0,0 +1,224 @@
+# Issue #47 — humanoid bulky/flat rendering: GfxObj close-degrade fix
+
+**Status:** root cause identified and patched (2026-05-06).
+**Flag:** `ACDREAM_RETAIL_CLOSE_DEGRADES=1` enables; off by default
+while the fix bakes.
+**Files:** `src/AcDream.Core/Meshing/GfxObjDegradeResolver.cs`,
+wiring in `src/AcDream.App/Rendering/GameWindow.cs`.
+
+## The bug, in one sentence
+
+acdream and ACViewer both rendered humanoid body parts using the
+**low-detail** GfxObj that the Setup / `AnimPartChange` references,
+instead of walking that base GfxObj's `DIDDegrade` table to slot 0
+(the close-detail mesh) the way retail does.
+
+## How we got here
+
+We spent a session investigating Issue #47 ("humanoid Setup 0x02000001
+renders bulky vs retail") and ruled out, with screenshots, every
+hypothesis on the original handoff list:
+
+- per-face vs smoothed vertex normals (smooth-normal pass had no
+  visible effect; dat normals were already smooth)
+- transform composition (acdream's `Scale * RotPart * TransPart *
+  RotEntity * TransEntity` matches retail's `Frame::combine` at
+  `0x518FD0` algebraically)
+- ambient floor / cell ambient tuning (lighting tweak, doesn't change
+  silhouette)
+- MSAA (anti-aliasing doesn't change silhouette thickness)
+- `client_highres.dat` precedence (retail does prefer HighRes over
+  Portal in `CLCache::GetDiskController` at `0x4f8fa0`, but the
+  humanoid body GfxObjs we were drawing don't get high-res
+  replacements — they get LOD replacements via DIDDegrade)
+- the `0x010001EC` null-part stubs in slots 17-33 (correctly skipped
+  per ACE's "essentially a null part" comment, but they were 1-tri
+  meshes — visually negligible, not the bug)
+
+The user critically reported that **ACViewer showed the same flat
+arms**, which meant the bug couldn't be in our renderer alone — it
+had to be in something both renderers shared. Both load from the same
+dat. Both run the AnimPartChange ids through their renderers as
+final mesh ids. Neither walks DIDDegrade.
+
+A side-by-side screenshot pair of `+Acdream` in retail vs acdream
+made the symptom precise: retail showed clear per-face linear gradients
+with visible bicep / deltoid / pectoral edges; acdream showed a smooth
+featureless tube.
+
+## Why retail looks different
+
+Retail's CPhysicsPart load and draw flow walks the degrade table:
+
+| Function                                | Address     | What it does                                                                                |
+|-----------------------------------------|-------------|---------------------------------------------------------------------------------------------|
+| `CPhysicsPart::LoadGfxObjArray`         | `0x0050DCF0`| Loads the base GfxObj only to read `DIDDegrade`. If a `GfxObjDegradeInfo` exists, retail loads each entry from `Degrades` into the part's render array. |
+| `CPhysicsPart::UpdateViewerDistance`    | `0x0050E030`| Picks `deg_level` per part by camera distance. For close / player rendering `deg_level == 0`. |
+| `CPhysicsPart::Draw`                    | `0x0050D7A0`| Draws `gfxobj[deg_level]`.                                                                  |
+
+So for close / player rendering the actual mesh is
+`GfxObjDegradeInfo.Degrades[0].Id`, NOT the base GfxObj id.
+
+## Concrete evidence
+
+Comparing the base meshes the server hands us (post-AnimPartChange)
+against the close-detail meshes their `DIDDegrade` tables point at:
+
+| Body part                       | Base id        | Base verts/polys | Degrade table | Slot 0 close id | Close verts/polys |
+|---------------------------------|----------------|------------------|---------------|-----------------|-------------------|
+| Aluvian Male upper arm          | `0x01000055`   | 14 / 17          | `0x110006D0`  | `0x01001795`    | 32 / 60           |
+| Aluvian Male lower arm          | `0x01000056`   | 8 / 6            | (per dat)     | `0x0100178F`    | 22 / 39           |
+| Heritage variant upper arm      | `0x010004BF`   | (low)            | (per dat)     | `0x010017A8`    | (high)            |
+| Heritage variant lower arm-A    | `0x010004BD`   | (low)            | (per dat)     | `0x010017A7`    | (high)            |
+| Heritage variant lower arm-B    | `0x010004B7`   | (low)            | (per dat)     | `0x0100179A`    | (high)            |
+
+Drawing the base ids gave us visibly LOD-3 bodies on close-up players —
+no bicep, no deltoid contour, no shoulder geometry. The degrade-slot-0
+meshes have the geometry that produces the per-face gradients the user
+expected.
+
+## Pseudocode
+
+```
+TryResolveCloseGfxObj(getGfxObj, getDegradeInfo, gfxObjId)
+  → resolvedId, resolvedGfxObj
+
+  base = getGfxObj(gfxObjId)
+  if base is null:
+    return (gfxObjId, null, false)            # caller drops the part
+
+  resolved = (gfxObjId, base)
+
+  if base.Flags HasDIDDegrade is clear OR base.DIDDegrade == 0:
+    return (resolved, true)
+
+  info = getDegradeInfo(base.DIDDegrade)
+  if info is null OR info.Degrades is empty:
+    return (resolved, true)
+
+  closeId = info.Degrades[0].Id
+  if closeId == 0:
+    return (resolved, true)
+
+  closeObj = getGfxObj(closeId)
+  if closeObj is null:
+    return (resolved, true)
+
+  return ((closeId, closeObj), true)
+```
+
+Every fallback leaves the base mesh selected — better to render the
+low-detail variant than nothing at all when the dat is partial.
+
+## Wiring in `GameWindow.OnLiveEntitySpawnedLocked`
+
+The order matters because the resolver has to see the **final**
+per-part GfxObj id, and downstream consumers (texture-change
+resolution, palette detection, mesh build) have to see the resolved
+mesh's surfaces:
+
+```
+1. Setup flatten → per-part transforms with default GfxObj ids.
+2. Apply server AnimPartChanges → replace per-part ids with the
+   body / clothing / head GfxObjs the server picked.
+3. *** NEW ***  If retail close-degrades enabled AND the setup is a
+   humanoid (34 parts with ≥8 null-sentinel slots in 17–33), run
+   each part's id through GfxObjDegradeResolver and swap to slot 0.
+4. Resolve TextureChanges against the resolved GfxObj's surfaces.
+5. Build palette overrides.
+6. GfxObjMesh.Build / texture upload.
+```
+
+Wiring it before AnimPartChanges would replace Setup defaults that
+will get overwritten anyway. Wiring it after texture-change resolution
+would point texture overrides at the wrong surface ids.
+
+## Scope
+
+For now the swap is gated to humanoid setups only. The detector is
+purely structural: 34 parts with at least 8 of slots 17-33 wired to
+the AC null-part sentinel `0x010001EC`. This matches Aluvian Male
+(`0x02000001`), the heritage variants, and any future 34-part
+humanoid sibling without enumerating ids.
+
+Why scoped vs. always-on:
+
+- Scenery and creatures may have degrade tables too (buildings
+  certainly do). For non-humanoids we haven't visually verified
+  that swapping to slot 0 is correct for the current camera distance,
+  so we hold the change.
+- True LOD plumbing (distance-based `deg_level` selection per
+  `CPhysicsPart::UpdateViewerDistance`) is still future work; until
+  then "always slot 0" is right for player + nearby NPCs but might
+  over-detail far-distance scenery.
+
+When the close-degrade path is validated everywhere, drop the
+humanoid scoping and remove the env-var flag.
+
+## Verification
+
+```powershell
+# Acceptance: side-by-side screenshots of `+Acdream` (or any humanoid
+# NPC) in acdream vs retail show matching shoulder / bicep / back
+# definition. Drudges and other monster setups stay correct.
+
+Get-Process -Name AcDream.App -ErrorAction SilentlyContinue | Stop-Process -Force
+Start-Sleep -Seconds 4
+
+$env:ACDREAM_RETAIL_CLOSE_DEGRADES = "1"
+$env:ACDREAM_DUMP_CLOTHING         = "1"   # log resolver swaps per spawn
+$env:ACDREAM_DAT_DIR   = "$env:USERPROFILE\Documents\Asheron's Call"
+$env:ACDREAM_LIVE      = "1"
+$env:ACDREAM_TEST_HOST = "127.0.0.1"
+$env:ACDREAM_TEST_PORT = "9000"
+$env:ACDREAM_TEST_USER = "testaccount"
+$env:ACDREAM_TEST_PASS = "testpassword"
+dotnet run --project src\AcDream.App\AcDream.App.csproj --no-build -c Debug 2>&1 |
+    Tee-Object -FilePath "launch_issue47_close_degrade.log"
+```
+
+Expected log lines (per spawn):
+
+```
+DEGRADE part=01 gfx=0x0100004F -> close=0x0100178D
+DEGRADE part=02 gfx=0x0100004D -> close=0x01001787
+DEGRADE part=10 gfx=0x0100122B -> close=0x01001795
+…
+```
+
+(Exact ids vary by which body parts AnimPartChange installs for the
+character's heritage / equipped clothing. The `->` arrow confirms
+the swap fired.)
+
+## What was rejected
+
+These were diagnostic experiments during the investigation, NOT part
+of the fix:
+
+- Smooth-normal recompute behind `ACDREAM_SMOOTH_NORMALS`
+- HighRes-first lookup in `TextureCache.DecodeFromDats`
+- Skipping `0x010001EC` null-part placeholders
+- Per-vertex Gouraud shader rewrite of `mesh.frag`
+- Cell ambient floor / minimum diffuse tuning
+- MSAA toggle
+- Identity per-part orientation
+- Positive-only polygon emission
+
+The successful fix is ONLY the close GfxObj degrade slot 0 swap. All
+of the above were reverted before this patch landed.
+
+## References
+
+- `acclient!CPhysicsPart::LoadGfxObjArray` at `0x0050DCF0` —
+  `docs/research/named-retail/acclient_2013_pseudo_c.txt`
+- `acclient!CPhysicsPart::UpdateViewerDistance` at `0x0050E030`
+- `acclient!CPhysicsPart::Draw` at `0x0050D7A0`
+- DatReaderWriter:
+  `references/DatReaderWriter/DatReaderWriter/Generated/DBObjs/GfxObj.generated.cs`
+  (`HasDIDDegrade` flag, `DIDDegrade` field)
+  `references/DatReaderWriter/DatReaderWriter/Generated/DBObjs/GfxObjDegradeInfo.generated.cs`
+  (`Degrades : List<GfxObjInfo>`)
+- ACE: `references/ACE/Source/ACE.DatLoader/FileTypes/GfxObjDegradeInfo.cs`
+- ACViewer + ACME both miss this same step — they draw the base id
+  directly. ACViewer's confirmation was the key signal that the bug
+  isn't acdream-specific.

docs/research/2026-05-06-issue-47-handoff.md

@@ -0,0 +1,287 @@
+# Issue #47 handoff — humanoid Setup 0x02000001 renders bulky vs retail
+
+**Use this whole document as the prompt** when handing off to a fresh agent.
+Everything they need to pick up cold is below.
+
+---
+
+## The bug, in one paragraph
+
+acdream renders any character that uses Setup `0x02000001` (Aluvian Male)
+with a visibly **bulkier, less shape-defined silhouette** than the same
+character viewed in retail's AC client. Specifically: shoulders look
+smoother / rounder where retail has pointier shoulder pads; the back lacks
+contour ("flat back"); arms appear puffier. The bug applies equally to
+players (`+Acdream`, `+Je`) and humanoid NPCs using the same setup
+(`Woodsman`, `Sedor Wystan the Blacksmith`, `Thelnoth Cort the Healer`,
+others). It is **independent of equipment** — `+Je` stripped naked still
+shows the bulky shape. It is **specific to Setup 0x02000001** — drudges
+(setup `0x020007DD`) and other monster setups render identically to retail
+through the same pipeline. See ISSUES.md `#47` for the full filing.
+
+## Acceptance criterion
+
+Side-by-side screenshots of the same humanoid (player or NPC) viewed from
+the same approximate angle in acdream and retail show **matching silhouette
+and shape definition** — pointy shoulders where retail has them, contoured
+back, no "puffy" arms. User confirms visually. NPCs, drudges, and scenery
+must continue to render correctly (no regressions).
+
+## What's already been ruled out (don't redo these)
+
+1. **0xF625 ObjDescEvent appearance updates being dropped.** Was a real
+   bug for skin/hair colors. Fixed in commit `e471527`. Does NOT affect
+   the bulky-shape issue (persists with the fix in place AND with no
+   equipment).
+2. **Position-pop on equip toggle.** Side effect of the appearance fix,
+   also resolved in `e471527`. Doesn't affect shape.
+3. **Clothing/armor overlapping the base body** (the HiddenParts
+   hypothesis). User stripped `+Je` naked; bulky shape persists.
+4. **`ParentIndex` hierarchy not walked in `SetupMesh.Flatten`.** Setup
+   `0x02000001` has a real hierarchy
+   (`-1, -1, 1, 2, 3, -1, 5, 6, 7, 0, 9, 10, 11, 12, 13, 14, 15, 0, ...`)
+   but a parent-walk implementation produced **no visible change**.
+   Confirms AC's idle animation frames are already in setup-root
+   coordinates, not parent-local.
+5. **Equipment / wielded items.** No equipment on `+Je` and bug persists.
+6. **Player-specific data flow.** Humanoid NPCs using the same setup
+   (Woodsman et al) show the same bug.
+7. **Silent GfxObj load failures or polygon drops.** `ACDREAM_DUMP_CLOTHING=1`
+   confirmed every one of the 34 parts emits triangles (`EMIT part=NN
+   gfx=0xXX subMeshes=N tris=N`); total ~648-700 tris per character.
+
+## What's confirmed (use this as starting facts)
+
+- `Setup.Parts.Count = 34` for `0x02000001`. `flatten.Count = 34`.
+  `AnimPartChanges = 34..38` depending on equipment. All match.
+- Idle animation frames place parts at sensible humanoid Z-heights
+  (`head Z=1.587, mid-body Z=0.5-1.0, ground Z=0.085`).
+- All 34 per-part orientations are nearly identical: 180° around -Z
+  axis (`W≈0, Z≈-1`). This is a setup-wide coordinate-flip convention.
+  Drudges have varied per-part orientations — different layout.
+- `setup.DefaultScale.Count = 0` for both humans and drudges → all parts
+  use `Vector3.One` scale.
+- Same fragment shader (`mesh.frag`) is used for humans and drudges.
+  Per-pixel diffuse with interpolated `vWorldNormal`.
+
+## Top hypotheses, ordered by likely payoff
+
+### Hypothesis A — per-face vs smoothed vertex normals
+
+**Strongest candidate.** AC's dat stores ONE normal per `SWVertex`. If
+human-character GfxObjs (e.g. `0x01001212`, `0x0100004B`-`0x01000059`)
+were authored with **per-face flat normals** (each vertex's normal copied
+from its triangle's face normal) while monster GfxObjs were authored with
+**smoothed normals** (averaged across adjacent faces), acdream's
+`Vector3.Normalize(sw.Normal)` would produce flat shading on humans and
+smooth shading on monsters. The screenshots strongly support this — retail
+characters look smooth-shaded, acdream characters look facet-edged.
+
+User said "not shaders" but they may not realize per-vertex normal
+*style* is part of the shader pipeline.
+
+**Test:** in `src/AcDream.Core/Meshing/GfxObjMesh.cs:142`, replace the
+direct `sw.Normal` read with a smooth normal computed per-load via
+face-adjacency accumulation:
+
+```csharp
+// Pre-pass: for each polygon, compute face normal; accumulate onto each vertex.
+// Post-pass: normalize.
+```
+
+Verify retail does this — see `docs/research/deepdives/r13-dynamic-lighting.md:107`
+for the `v.normal += t.face_normal` pattern, and
+`docs/plans/2026-04-13-rendering-rebuild.md:50` for the
+`AdjustPlanes: face-normal accumulation + per-vertex lighting` note (terrain context but
+same math applies to characters).
+
+If smooth-normals fixes humans and ALSO doesn't break drudges (because
+drudge dat normals were already smooth, computing them again gives the
+same answer modulo precision), this is the bug.
+
+### Hypothesis B — cell ambient too low
+
+Back-facing surfaces (the parts of the character not lit by the directional
+sun) fall to `uCellAmbient` in `mesh.frag`. If ambient is very dark, the
+back of any character looks uniformly black — reading as "flat" because no
+detail variation is visible. Retail likely has higher ambient that lets
+unlit surfaces still show their geometry through subtle gradients.
+
+**Test:** dump `uCellAmbient` UBO values during a player render and compare
+to retail's behaviour. Try bumping ambient temporarily and see if back-
+detail emerges.
+
+### Hypothesis C — anti-aliasing
+
+acdream's GL window may not have MSAA enabled. Without it, polygon edges
+visibly stair-step, exaggerating the faceted look at low triangle counts
+(~700 tris per character). Retail likely has AA on by default.
+
+**Test:** check the Silk.NET window creation code for `Samples`/MSAA
+config. Try enabling `Samples=4` and re-render.
+
+### Hypothesis D — orientation composition order or sign
+
+The 180°-around-Z rotation on every part is unusual. If acdream applies
+it correctly but retail applies it differently (e.g. as a post-multiply
+or with the inverse), parts could be subtly mis-positioned in ways that
+read as "bulky" rather than "broken". My investigation didn't fully rule
+this out — `parent-walk` was a no-op, but a *single-level* orientation
+composition discrepancy might be invisible without comparing actual
+post-transform vertex positions to retail.
+
+**Test:** attach cdb to retail (see CLAUDE.md "Retail debugger toolchain"),
+break in `Frame::combine` (`0x518FD0`) with a player guid, dump the
+resulting `Frame` for parts 0, 9, 16. Compare to acdream's per-part
+world matrices (add a diagnostic).
+
+## Diagnostic infrastructure already built
+
+All env-var-gated, no runtime cost when off:
+
+```powershell
+$env:ACDREAM_DUMP_CLOTHING = "1"
+```
+
+Prints, for every humanoid spawn (gate: `setup.Parts.Count >= 10`):
+
+- Header: `setup.Parts.Count`, `flatten.Count`, `APC` count
+- `ParentIndex[N]: -1,-1,1,2,3,...` array
+- `DefaultScale[N]: ...` array
+- `IdleFrame.Frames[N]:` per-part `Origin` + `Orientation` (first 17 parts)
+- `EMIT part=NN gfx=0xXXXXXXXX subMeshes=N tris=N` per part
+- `TOTAL tris=N meshRefs=N` per entity
+
+```powershell
+$env:ACDREAM_DUMP_APPEARANCE = "1"
+```
+
+Prints structured decode of every `0xF625 ObjDescEvent` (mid-session
+appearance change) — SubPalettes, TextureChanges, AnimPartChanges.
+
+Source: `src/AcDream.App/Rendering/GameWindow.cs` around `dumpClothing` /
+`OnLiveEntitySpawnedLocked`, and `src/AcDream.Core.Net/WorldSession.cs`
+`DumpAppearanceEnabled`.
+
+## Workflow (per `CLAUDE.md`)
+
+This is AC-specific behavior. Follow the mandatory workflow:
+
+1. **Step 0 — grep named first.**
+   `grep "Class::method" docs/research/named-retail/acclient_2013_pseudo_c.txt`
+   for any function name from a hypothesis. The Sept 2013 PDB is named —
+   most things have real names. Don't decompile fresh until you've grep'd.
+2. **Step 1 — cross-reference.** Check ACE / ACME / ACViewer / Chorizite /
+   AC2D / holtburger as appropriate. The reference hierarchy in CLAUDE.md
+   spells out which ref is authoritative for each domain. For rendering:
+   ACME `WorldBuilder.Tests/ClientReference.cs` is the closest oracle.
+3. **Step 2 — pseudocode.** Write the algorithm in plain language under
+   `docs/research/*_pseudocode.md` before porting.
+4. **Step 3 — port faithfully.** Match retail line-by-line, same variable
+   names, same control flow. No "improvements".
+5. **Step 4 — conformance test.** Write a test using a captured wire body
+   or known dat values as golden output.
+6. **Step 5 — visual verification.** User confirms in-client.
+
+If grep-named-first turns up nothing relevant, the function is likely
+in the unnamed minority. Fall back to Ghidra chunks under
+`docs/research/decompiled/` keyed by address.
+
+## Files most likely to need edits
+
+- `src/AcDream.Core/Meshing/GfxObjMesh.cs` — polygon emission, vertex
+  normal handling at line 142. **Hypothesis A lives here.**
+- `src/AcDream.Core/Meshing/SetupMesh.cs` — per-part transform
+  composition. The parent-walk experiment lives here (and was reverted —
+  see git history if you want to revisit it).
+- `src/AcDream.App/Rendering/Shaders/mesh.frag` — per-pixel lighting
+  equation. **Hypothesis B lives here** if ambient is the cause.
+- `src/AcDream.App/Rendering/Shaders/mesh.vert` — normal transform
+  (`mat3(uModel) * aNormal`). Watch for non-uniform scale issues if
+  any future change touches scale.
+- The Silk.NET window setup code (search for `IWindow.Create` /
+  `WindowOptions`) — **Hypothesis C lives here** if MSAA needs enabling.
+
+## Test workflow (live verification)
+
+```powershell
+# Always: kill stale + 3-5s wait (ACE session lingers briefly).
+Get-Process -Name AcDream.App -ErrorAction SilentlyContinue | Stop-Process -Force
+Start-Sleep -Seconds 4
+
+$env:ACDREAM_DUMP_CLOTHING = "1"  # verbose per-spawn diagnostics
+$env:ACDREAM_DAT_DIR   = "$env:USERPROFILE\Documents\Asheron's Call"
+$env:ACDREAM_LIVE      = "1"
+$env:ACDREAM_TEST_HOST = "127.0.0.1"
+$env:ACDREAM_TEST_PORT = "9000"
+$env:ACDREAM_TEST_USER = "testaccount"
+$env:ACDREAM_TEST_PASS = "testpassword"
+dotnet run --project src\AcDream.App\AcDream.App.csproj --no-build -c Debug 2>&1 |
+    Tee-Object -FilePath "launch_issue47.log"
+```
+
+The user's `+Acdream` character logs in at Holtburg. Nearby:
+- Other player(s) like `+Je` (driven from a parallel retail acclient)
+- Humanoid NPCs (Woodsman, Sedor Wystan the Blacksmith, Thelnoth Cort
+  the Healer, Sontella Dagroff the Bowyer, Archmage Cindrue, Monyra the
+  Jeweler, Ecutha the Tailor) — most use Setup 0x02000001
+- Drudges and slinkers nearby (different setup, render correctly — use
+  as control)
+- The Foundry's "Nullified Statue of a Drudge" — different setup, used
+  for prior diagnostic work
+
+User has the retail acclient open in parallel for side-by-side comparison.
+Visual verification is the acceptance test — there's no automated way to
+say "this looks like retail."
+
+## Constraints / don't-break
+
+- Do **not** break drudge / monster rendering. They're correct now.
+- Do **not** break scenery (terrain, buildings, statues). They're correct.
+- Do **not** break the local `+Acdream` (own-character) view either —
+  same pipeline so any rendering change applies to it.
+- Tests must stay green: `dotnet test AcDream.slnx` should report 0
+  failures. There are 8 pre-existing motion test failures in
+  `AcDream.Core.Tests` that are NOT yours — don't try to fix them in
+  this work.
+- Build must stay green: `dotnet build AcDream.slnx -c Debug`.
+
+## When to stop and ask
+
+Per CLAUDE.md, ask only for:
+- Visual verification (user looking at the client)
+- Genuine architectural disagreements with the roadmap
+- Hard-to-reverse destructive actions
+
+Otherwise act. Don't ask "should I continue?".
+
+## References to consult
+
+- `references/ACME/WorldBuilder/Editors/Landscape/StaticObjectManager.cs` —
+  ACME's mesh hydration; same stack (Silk.NET) as us, used as our
+  closest "should look right" oracle.
+- `references/ACViewer/ACViewer/Physics/PartArray.cs:614` (`UpdateParts`)
+  and `references/ACViewer/ACViewer/Physics/Animation/AFrame.cs:43`
+  (`Combine`) — frame composition math.
+- `references/ACE/Source/ACE.Server/WorldObjects/WorldObject_Networking.cs:48`
+  (`SerializeUpdateModelData`) and `:978` (`AddBaseModelData`) — what
+  ACE puts on the wire for character appearance.
+- `docs/research/named-retail/acclient_2013_pseudo_c.txt` — Sept 2013 PDB
+  with 18,366 named functions. **Grep this FIRST.**
+- `docs/plans/2026-04-13-rendering-rebuild.md` — earlier rendering
+  rebuild plan with notes on `AdjustPlanes` (terrain, but the
+  face-normal accumulation pattern is the smooth-normal pattern).
+- `docs/research/deepdives/r13-dynamic-lighting.md` — has the smooth-
+  normal accumulation pseudocode at line 107.
+
+## Final note
+
+This is a **rendering-fidelity** issue, not a wire-protocol one. The
+network data is correct (the previous session's `0xF625` work confirmed
+it). The bug is in how acdream interprets that data into pixels.
+
+The smoothest path is probably Hypothesis A (smooth normals), one
+contained change in `GfxObjMesh.Build`, gated behind a feature flag for
+A/B testing, then verified live by the user. If that doesn't fix it,
+move to B (ambient), then C (MSAA), then D (frame composition with cdb
+trace).

docs/research/2026-05-06-locomotion-cycle-transitions/findings-static.md

@@ -0,0 +1,252 @@
+# Locomotion-cycle transitions — static-analysis findings
+
+**Date:** 2026-05-06 (follow-up to `investigation-prompt.md`)
+**Status:** static analysis complete; **candidate fix INEFFECTIVE per visual verify** — see "Visual-verify result" at bottom; root cause is the same as 2026-05-03 hypothesis F (cycle stuck downstream of `Sequencer.CurrentMotion`).
+
+This is what code reading + ACE cross-reference + named-retail grep
+established before any TTD/cdb trace was run. It scopes the candidate
+fix and identifies which questions still need the trace.
+
+---
+
+## What's confirmed by static analysis
+
+### 1. ACE broadcasts unconditionally on inbound MoveToState
+
+`references/ACE/Source/ACE.Server/Network/GameAction/Actions/GameActionMoveToState.cs:36`
+calls `session.Player.BroadcastMovement(moveToState)` for every received
+MoveToState, no diff-check. So **whether ACE broadcasts a UM is determined
+purely by whether the client sent a MoveToState**.
+
+### 2. ACE auto-upgrades `WalkForward + HoldKey.Run → RunForward` on broadcast
+
+`references/ACE/Source/ACE.Server/Network/Motion/MovementData.cs:110-113`
+shows the conversion: when client sends `WalkForward + HoldKey.Run`, the
+broadcast `interpState.ForwardCommand` is upgraded to `RunForward`. So
+the broadcast UM byte differs between Run and Walk even though the
+client's wire byte is the same `WalkForward`.
+
+### 3. Therefore the bug is one of these two
+
+Either:
+- **A.** Retail (the actor) does NOT send a fresh MoveToState when only
+  HoldKey changes (Shift toggle while a direction key is held). ACE has
+  nothing to broadcast → no UM arrives at the observer. UPs continue
+  with new velocity (run-pace ↔ walk-pace), but observer's UM-driven
+  cycle never updates.
+- **B.** Retail DOES send a fresh MoveToState. ACE broadcasts a UM. Our
+  parser receives it but fails to apply the cycle change for some
+  reason.
+
+The diagnostic data captured in 2026-05-03's investigation
+(`[FWD_WIRE]`, `[UM_RAW]`, `[SETCYCLE]`) is more consistent with **A**:
+no `[FWD_WIRE]` line was logged for Shift toggles, suggesting no UM
+arrived. But that data was for a different scenario, so a fresh trace
+is needed.
+
+**This question is the primary gap the TTD/cdb trace must close.**
+A 2-minute cdb session with a breakpoint counter on
+`acclient!CommandInterpreter::SendMovementEvent` answers it
+definitively.
+
+### 4. `ApplyServerControlledVelocityCycle` was gated against player remotes at three sites
+
+In `src/AcDream.App/Rendering/GameWindow.cs` before this change:
+
+- **Inner gate**, function entry (~line 3326):
+  `if (IsPlayerGuid(serverGuid)) return;`
+- **Outer gate**, call site in `OnLivePositionUpdated` (~line 3683):
+  `if (!IsPlayerGuid(update.Guid) && rmState.HasServerVelocity ...)`
+- **TickAnimations gate** (~line 6325):
+  `if (!IsPlayerGuid(serverGuid) && rm.HasServerVelocity)` —
+  this one is for the "stale velocity → reset to zero" path, used by
+  NPCs whose MoveTo has expired. Leave as-is for now; player remotes
+  don't go through this branch (no MoveTo path).
+
+### 5. `ServerControlledLocomotion.PlanFromVelocity` thresholds are wrong for player speeds
+
+In `src/AcDream.Core/Physics/ServerControlledLocomotion.cs:30-33`:
+
+```csharp
+public const float StopSpeed = 0.20f;
+public const float RunThreshold = 1.25f;
+```
+
+For player speeds (Walk ≈ 2.5 m/s, Run ≈ 9 m/s — both observed in
+prior `[VEL_DIAG]` traces), `RunThreshold = 1.25` is below both
+buckets. Both speeds resolve to `RunForward`. So even if we just
+un-gated the existing function for player remotes, the function would
+*always* return `RunForward` and the bug would persist.
+
+The function works correctly for NPCs (typical NPC speed range
+1–6 m/s falls on the right side of the 1.25 threshold for the
+Idle/Walk/Run distinction).
+
+**Implication:** the player path must use different thresholds. The
+candidate fix introduces hysteresis tuned for player speeds (4.5 demote,
+5.5 promote) and routes player remotes through a dedicated
+`ApplyPlayerLocomotionRefinement` helper instead of `PlanFromVelocity`.
+
+### 6. `PlanFromVelocity` returns Forward-only motions
+
+It returns `Ready`, `WalkForward`, or `RunForward` — never sidestep,
+never backward. Even with thresholds fixed, calling it for a sidestepping
+player remote would clobber `SideStepLeft`/`Right` with `RunForward`.
+
+**Implication:** the player path must scope itself to forward direction
+only; sidestep/backward HoldKey toggles are deferred until TTD confirms
+retail's per-direction algorithm.
+
+---
+
+## What the candidate fix does (and doesn't)
+
+**Does:**
+
+- Adds `RemoteMotion.LastUMTime` (timestamp of most recent UM for a remote).
+- Stamps `LastUMTime` at the top of `OnLiveMotionUpdated`.
+- Removes the `IsPlayerGuid` early return inside
+  `ApplyServerControlledVelocityCycle`.
+- Removes the `!IsPlayerGuid` gate at the outer call site in
+  `OnLivePositionUpdated` (so player remotes get the function called).
+- Routes player remotes through new `ApplyPlayerLocomotionRefinement`:
+  - 500 ms UM grace window (UMs are authoritative).
+  - Forward-direction-only refinement (low byte 0x05 / 0x07).
+  - Hysteresis: Run → Walk demote at < 4.5 m/s; Walk → Run promote at
+    > 5.5 m/s.
+  - SetCycle skipped if neither motion ID nor speedMod changed
+    meaningfully.
+  - Diagnostic `[UPCYCLE_PLAYER]` line gated on `ACDREAM_REMOTE_VEL_DIAG=1`.
+
+**Does NOT (deferred to TTD-validated follow-up):**
+
+- Backward (case #2) HoldKey toggle.
+- Sidestep speed-bucket refinement (cases #4, #5).
+- Direction-flip transitions (cases #3, #6, #7) — these are believed
+  to work today via UM-driven SetCycle, but the prompt's acceptance
+  criteria explicitly call for visual confirmation of all 7 cases.
+- Tuning the grace window or the hysteresis thresholds against retail's
+  exact behavior. 500 ms / 4.5 / 5.5 are defensible defaults but TTD
+  may show retail uses different values.
+
+---
+
+## Acceptance for the candidate fix (case #1 only)
+
+When acdream observes a retail-driven character:
+
+- Hold W (run) for 2 s, then press Shift, then release Shift, then
+  release W. The visible leg cycle should switch Run → Walk → Run
+  → Idle within ~200–500 ms of each transition.
+- All other working cases (acdream-on-acdream, retail-on-acdream,
+  Idle ↔ Run, Idle ↔ Walk) must still work — no regressions.
+
+If the candidate fix produces visible Run↔Walk transitions on retail
+actors, ship it (close part of #39, file the remaining 6 cases as a
+follow-up issue) and run TTD on cases #2–#7 next.
+
+If it doesn't switch the cycle (or thrashes / regresses something),
+the next investigation step is the cdb breakpoint count on
+`CommandInterpreter::SendMovementEvent` to settle question A vs. B
+above before any further fix attempt.
+
+---
+
+## Files touched by the candidate fix
+
+- `src/AcDream.App/Rendering/GameWindow.cs`
+  - Added `RemoteMotion.LastUMTime` field
+  - Stamped `LastUMTime` at top of `OnLiveMotionUpdated`
+  - Removed inner `IsPlayerGuid` gate in `ApplyServerControlledVelocityCycle`
+  - Routed player remotes to new `ApplyPlayerLocomotionRefinement`
+  - Added constants `UmGraceSeconds`, `PlayerRunPromoteSpeed`,
+    `PlayerRunDemoteSpeed`
+  - Removed `!IsPlayerGuid` gate at outer call site in `OnLivePositionUpdated`
+
+No changes to `ServerControlledLocomotion.cs`, `MotionInterpreter.cs`,
+or `AnimationSequencer.cs`.
+
+---
+
+## Visual-verify result (2026-05-06, evening)
+
+**The candidate fix did not switch the cycle.** User report: "if I shift
+walk and then release shift it just blips forward in walking animation."
+
+Diagnostic dump from `launch-39-candidate.log` (filtered to the
+retail-driven actor, guid `0x50000001`):
+
+| Diag | Result |
+|---|---|
+| `[FWD_WIRE]` | **Many direct Walk↔Run transitions present** — `oldCmd=0x44000007 → newCmd=0x45000005` and reverse. ACE IS broadcasting UMs on Shift toggle (refutes the static-analysis hypothesis that retail doesn't send MoveToState on HoldKey-only changes). |
+| `[UM_RAW]` `seq.CurrentMotion` distribution for `0x50000001` | 0x41000003: 63, 0x45000005: 52, 0x6500000D: 31, 0x44000007: 20, 0x6500000F: 6 — i.e., the sequencer's `CurrentMotion` field DOES change to Walk / Run / Turn / Sidestep at various sample moments. |
+| `[SETCYCLE]` | Many direct Walk↔Run SetCycle calls — `old=(motion=0x45000005…) new=(motion=0x44000007…)` — confirming the picker correctly resolves Run cycle from the inbound UM. |
+| `[SCFULL]` | After Walk→Run SetCycle: `currNodeIsCyclic=False`, `qCount=12, 13, 14, … 41, 42, 49`. Queue grows steadily over ~30 transitions; `_currNode` lands on a non-cyclic transition link rather than the new cycle. |
+| `[PARTSDIAG]` `seqHash` | 447 distinct values across 467 samples — animation IS being rendered, frames ARE changing. The cycle just isn't the right one. |
+| `[UPCYCLE_PLAYER]` | **Never fired.** The 500 ms UM grace window blocks every call (UMs arrive at >2 Hz, well within the grace), which is the correct behavior — UMs are authoritative when fresh. |
+
+### What this proves
+
+1. **Wire is correct.** ACE delivers UMs on Shift toggle (refutes
+   hypothesis A from this doc, refutes the original 2026-05-06
+   investigation-prompt's wire-level hypothesis).
+2. **Dispatch is correct.** `OnLiveMotionUpdated` parses each UM,
+   resolves the right cycle, calls `AnimationSequencer.SetCycle`.
+3. **Sequencer state updates.** `Sequencer.CurrentMotion` reaches the
+   new motion ID (Run, Walk, Turn, Sidestep all observed in `[UM_RAW]`).
+4. **Visible animation does NOT match `CurrentMotion`.** This is exactly
+   the same residual symptom the 2026-05-03 investigation chased and
+   the SetCycle force-`_currNode` fix in commit `357dcc0` attempted but
+   did not fully resolve.
+
+### Where the bug actually lives
+
+Downstream of `Sequencer.CurrentMotion`, in one of:
+
+- `AnimationSequencer.Advance(dt)` consuming the wrong queue node
+- `AnimationSequencer.BuildBlendedFrame()` reading from a stale
+  `_currNode`
+- The transition link from old-cycle to new-cycle never advancing to
+  the cycle (link plays as the old motion)
+- Queue-management gap: retail's `MotionTableManager::CheckForCompletedMotions` /
+  `remove_redundant_links` (named-retail addresses 0x0051BE00 / 0x0051BF20)
+  are not ported. Our `qCount=49` after ~30 transitions confirms
+  unbounded queue growth — but `_currNode` should still be on the most
+  recent SetCycle's first node, so this alone may not be the bug.
+
+The 2026-05-03 prompt's hypothesis F is the most likely:
+
+> `Advance` plays through stale link frames before reaching the cycle.
+> Our `357dcc0` fix forces `_currNode` onto the first newly-enqueued
+> node — but for `Walk→Run`, the newly-enqueued sequence is `[walk-to-run
+> link, run cycle]`. `_currNode` lands on the **link**, the link plays
+> for ~0.5–1 second, then the run cycle starts. User perceives the
+> link's "transition pose" as "still walking."
+
+The SCFULL data (`currNodeIsCyclic=False` after every Walk↔Run direct
+transition) is consistent with this — `_currNode` is on the link,
+not the cycle, immediately after SetCycle.
+
+### What to do next
+
+The candidate fix code (`LastUMTime`, `ApplyPlayerLocomotionRefinement`,
+hysteresis constants, un-gated call site) is **harmless but ineffective**
+— left in place because:
+- The infrastructure may be useful for cases #2–#7 if/when those turn
+  out to need velocity fallback.
+- The UM grace window is correctly preventing it from interfering with
+  the working UM-driven path.
+- Reverting would lose context tied to the findings doc.
+
+The next investigation should target **AnimationSequencer.Advance / queue
+management for cyclic→cyclic direct transitions** — this is the same
+problem 2026-05-03 left open. Recommended path:
+
+1. Add per-tick diag for `_currNode.Anim.Id` + `_framePosition` (per
+   the 2026-05-03 prompt's "Concrete next steps" §1).
+2. Cross-reference retail's `CSequence::update` via cdb live attach OR
+   read it in named-retail (it's named — search `acclient_2013_pseudo_c.txt`
+   for `CSequence::update_internal`).
+3. Compare retail's queue-handling for direct cyclic→cyclic with our
+   `AnimationSequencer.Advance` — particularly the conditions under which
+   `_currNode` advances past a link to the next cycle.

docs/research/2026-05-06-locomotion-cycle-transitions/investigation-prompt.md

@@ -0,0 +1,305 @@
+# Locomotion-cycle transitions on observed remotes — investigation prompt
+
+**Hand-off date:** 2026-05-06
+**Status:** open. ISSUES.md #39 captured the Run↔Walk-forward variant; this
+prompt expands to the full locomotion transition matrix and lays out a
+TTD-driven investigation against retail to ground the fix.
+
+This document is a self-contained briefing for an agent (or fresh session)
+picking this up. The TTD toolchain landed in commit `e3e5bf5` (#44) and
+is the primary investigative tool here.
+
+---
+
+## What problem are we trying to solve?
+
+When acdream observes a remote-driven player character (typically a parallel
+retail acclient.exe connected to the same local ACE), the visible leg-cycle
+animation does not always switch when the actor changes locomotion mode.
+The body translates at the right speed (server-driven velocity is fine), but
+the legs keep playing whichever cycle was active before.
+
+The full set of transitions where the bug may surface:
+
+| # | Transition | Wire change | Likely cause |
+|---|---|---|---|
+| 1 | Run forward ↔ Walk forward (Shift toggle while W held) | HoldKey + ForwardSpeed only | ACE doesn't broadcast UM (no ForwardCommand change); cycle refinement must come from UP-derived velocity |
+| 2 | Run backward ↔ Walk backward (Shift toggle while S held) | HoldKey + ForwardSpeed only | Same as #1, backward axis |
+| 3 | Forward ↔ Backward (W→S or S→W direct flip) | ForwardCommand changes (e.g. `WalkForward` → `WalkBackward`) | ACE broadcasts a UM; cycle should update from UM directly |
+| 4 | Fast strafe-left ↔ Slow strafe-left (Shift toggle while A held) | HoldKey + SideSpeed only | Same as #1 / #2 — speed-bucket only |
+| 5 | Fast strafe-right ↔ Slow strafe-right (Shift toggle while D held) | HoldKey + SideSpeed only | Same |
+| 6 | Strafe-Left ↔ Strafe-Right (A↔D direct flip) | SideCommand changes | ACE broadcasts UM |
+| 7 | Forward ↔ Strafe (W↔A, W↔D, etc.) | ForwardCommand or SideCommand changes | UM broadcast |
+
+#39 in ISSUES.md focused on #1. This investigation widens scope to the
+whole matrix because the underlying wire pattern is the same shape: half
+the transitions broadcast a UM, the other half rely on UP-derived velocity.
+A correct fix should handle both classes uniformly.
+
+---
+
+## What we already know (don't re-discover)
+
+From the prior 2026-05-03 investigation (`docs/research/2026-05-03-remote-anim-cycle/`)
+and #39 in `docs/ISSUES.md`:
+
+- **ACE only broadcasts a fresh UpdateMotion when the wire's `ForwardCommand`
+  byte changes** — i.e. on direction-key state changes. Toggling Shift while
+  W held changes `ForwardSpeed` and `HoldKey` but NOT `ForwardCommand`, so
+  ACE does NOT broadcast a UM for the demote/promote.
+- **Speed changes still propagate via UpdatePosition (UP).** Position-delta
+  velocity changes between Run-pace and Walk-pace, confirmed via
+  `[VEL_DIAG]` serverSpeed varying ~2.5 m/s (walk) ↔ ~9 m/s (run).
+- **Retail's inbound code uses UP-derived velocity to refine the visible
+  cycle when no UM tells it.** Acdream has the equivalent function —
+  `ApplyServerControlledVelocityCycle` in `GameWindow.cs:3274` — but
+  it's gated `if (IsPlayerGuid(serverGuid)) return;` for player remotes,
+  exactly the case where the gap matters.
+- **The fix sketch is small (~10 lines):** un-gate
+  `ApplyServerControlledVelocityCycle` for player remotes when
+  `currentMotion` is a locomotion cycle. Add a `LastUMUpdateTime` grace
+  window so fresh UMs win over UP-velocity refinement.
+- **Outbound from acdream is fine** — the matrix in #39 confirms retail
+  observers see acdream's transitions correctly. The bug is purely in
+  acdream's RECEIVE path.
+
+What's NOT yet confirmed (and is the primary goal of this investigation):
+
+- **Which exact retail function** does the cycle refinement when no UM
+  arrives? Hypothesis: `CPhysicsObj::unpack_movement` →
+  `MovementManager::unpack_movement` → some velocity-aware cycle picker.
+  We need to identify the function name and what state it reads.
+- **What grace window** does retail use between UM and UP-derived
+  refinement? Our hypothesis is ~500 ms but it may be different.
+- **Does retail use SideSpeed or SideCommand changes** the same way for
+  strafe transitions?
+
+---
+
+## TTD-driven workflow
+
+The toolchain shipped in #44 (commit `e3e5bf5`). Bootstrap is one-time
+per machine — see CLAUDE.md "TTD recordings (offline replay)" or
+`tools/ttd-record.ps1` header.
+
+### Recording scenario
+
+You need **two retail acclient.exe instances** running on the same local
+ACE — one as the recorded process, one as the actor. Acdream can be
+involved as a third client for cross-checking but is not strictly required
+for the recording itself.
+
+Setup:
+1. Launch retail #1 on `testaccount` / `+Acdream` (or any character).
+2. Launch retail #2 on a second account / character. It must be on the
+   same landblock so retail #1 can see it.
+3. Position retail #2 in retail #1's view, ~10 m away.
+4. In an **elevated** PowerShell, attach TTD to retail #1 (the observer):
+   ```powershell
+   tools\ttd-record.ps1 -RingMaxMB 512
+   ```
+5. Wait for `Recording is now active`. Retail #1 will be ~10× slower —
+   that's expected.
+
+Scenario (drive on retail #2, ~60 seconds total):
+
+```
+Phase 1 — Forward speed:    Hold W (2 s) → +Shift → -Shift → +Shift → -Shift → release W (idle 1 s)
+Phase 2 — Backward speed:   Hold S (2 s) → +Shift → -Shift → +Shift → -Shift → release S (idle 1 s)
+Phase 3 — Forward↔Back flip: Hold W (2 s) → release W → Hold S (2 s) → release S
+Phase 4 — Strafe-Left speed: Hold A (2 s) → +Shift → -Shift → +Shift → -Shift → release A
+Phase 5 — Strafe-Right speed: Hold D (2 s) → +Shift → -Shift → +Shift → -Shift → release D
+Phase 6 — Strafe L↔R flip:  Hold A (2 s) → release A → Hold D (2 s) → release D
+Phase 7 — Forward↔Strafe:   Hold W (2 s) → release W → Hold D (2 s) → release D
+```
+
+After phase 7, Ctrl+C the TTD recording. Trace lands at
+`~/.ttd/traces/acclientNN.run`.
+
+### Query strategy
+
+The recording captured ~7 phases in ~60 sec. Each phase produced specific
+UM and/or UP traffic. Goal: for each phase identify what retail's receive
+code did to update the visible cycle.
+
+Start with a `.cdb` script under `tools/ttd-queries/locomotion.cdb`:
+
+```
+.echo === Top motion-receive entry points ===
+dx -r2 @$cursession.TTD.Calls("acclient!CPhysicsObj::unpack_movement").Take(20)
+dx -r2 @$cursession.TTD.Calls("acclient!MovementManager::unpack_movement").Take(20)
+
+.echo === Cycle-update functions called ===
+dx -r2 @$cursession.TTD.Calls("acclient!*set_animation*").GroupBy(c => c.Function).Select(g => new { Name = g.First().Function, Count = g.Count() }).OrderByDescending(x => x.Count)
+dx -r2 @$cursession.TTD.Calls("acclient!*SetCycle*").GroupBy(c => c.Function).Select(g => new { Name = g.First().Function, Count = g.Count() }).OrderByDescending(x => x.Count)
+dx -r2 @$cursession.TTD.Calls("acclient!*play_*").GroupBy(c => c.Function).Select(g => new { Name = g.First().Function, Count = g.Count() }).OrderByDescending(x => x.Count)
+
+.echo === Velocity / position-update handlers ===
+dx -r2 @$cursession.TTD.Calls("acclient!*UpdatePosition*").GroupBy(c => c.Function).Select(g => new { Name = g.First().Function, Count = g.Count() }).OrderByDescending(x => x.Count)
+dx -r2 @$cursession.TTD.Calls("acclient!CPhysicsObj::set_velocity").Take(15)
+
+.echo === DONE ===
+q
+```
+
+Run it:
+```powershell
+tools\ttd-query.ps1 -Script tools\ttd-queries\locomotion.cdb
+```
+
+**Then iterate.** Use the function-name hits to guide the next query —
+inspect call args (`Take(N)` instead of `Count()`), navigate to specific
+TTD positions and dump struct fields with `dt acclient!ClassName <addr>`.
+
+The named-retail decomp at `docs/research/named-retail/acclient_2013_pseudo_c.txt`
+has the source for every retail function. Grep it by class::method to
+read what the function does, then use TTD to confirm what it does at
+runtime.
+
+### Specifically what to look for
+
+For each phase in the recording, answer:
+
+1. **Did a UM arrive?** Look for calls to the receive entry points around
+   that phase's time window. UM dispatch should fire on direction-key
+   changes (phases 3, 6, 7) but NOT on Shift-only toggles (phases 1, 2, 4, 5).
+2. **What function refined the cycle when no UM arrived?** Hypothesis is
+   something in the velocity-aware path. Find the symbol, read the
+   decomp, document the conditions.
+3. **What threshold/grace logic is in place?** Retail must have some way
+   to prevent UP-velocity refinement from fighting a fresh UM. Usually
+   a timestamp comparison.
+
+---
+
+## The fix
+
+The fix lives in `src/AcDream.App/Rendering/GameWindow.cs:3274`
+(`ApplyServerControlledVelocityCycle`). Current code returns early for
+player remotes. The fix is to:
+
+1. **Un-gate the early return** when the current motion is a locomotion
+   cycle. Locomotion cycles to detect: `0x44000007` (Run forward),
+   `0x45000005` (Walk forward), backward equivalents, sidestep variants.
+   Keep the gate when the motion is non-locomotion (e.g. emotes, attacks).
+2. **Add a `LastUMUpdateTime` per remote.** Touch it in the UM handler
+   path. In `ApplyServerControlledVelocityCycle`, skip refinement if
+   `(now - LastUMUpdateTime) < graceMs`. Start with `graceMs = 500`
+   and tune against the TTD findings.
+3. **Use UP-derived velocity to pick the speed bucket.** Existing logic
+   in `ServerControlledLocomotion.PlanFromVelocity` already does this —
+   verify thresholds match retail's bucketing (TTD will tell you the
+   exact boundaries retail uses).
+
+For direction-flip transitions (phases 3, 6, 7), the UM handler path
+should already work — the visible cycle should update from the UM
+directly. Confirm in-client that those transitions are clean BEFORE
+shipping the velocity-fallback fix; if they're broken too, that's a
+separate UM-handler bug that needs its own investigation.
+
+---
+
+## Acceptance criteria
+
+In acdream observing a retail-driven character:
+
+- **Phases 1, 2, 4, 5 (Shift toggles):** visible leg cycle switches
+  Run↔Walk / Fast↔Slow within ~200 ms of the wire change.
+- **Phases 3, 6, 7 (direction flips):** visible cycle updates within
+  ~100 ms (UM is direct-path, faster than UP).
+- **No regression** on already-working cases:
+  - acdream-on-acdream (matrix row in #39)
+  - retail observers viewing acdream (works today)
+  - Idle ↔ Run transitions
+  - Idle ↔ Walk transitions
+  - Combat-mode locomotion (if testable)
+- **No spurious cycle thrashing during turning while running** —
+  ObservedOmega-driven body rotation must not trigger
+  velocity-bucket changes mid-cycle.
+
+---
+
+## Files to read first
+
+In this order:
+
+1. `docs/ISSUES.md` — issue #39 (Run↔Walk specific) and #44 (TTD toolchain)
+2. `docs/research/2026-05-03-remote-anim-cycle/investigation-prompt.md` —
+   prior investigation, what's been tried, what works
+3. `CLAUDE.md` — "Retail debugger toolchain" section (live cdb +
+   "TTD recordings (offline replay)" subsection)
+4. `memory/project_retail_debugger.md` — durable lessons + TTD addendum
+5. `memory/project_retail_motion_outbound.md` — what retail's outbound
+   motion path looks like (cdb live trace from 2026-05-01)
+
+Then:
+
+6. `src/AcDream.App/Rendering/GameWindow.cs` — `OnLiveMotionUpdated`
+   (~line 3203) and `ApplyServerControlledVelocityCycle` (~line 3274)
+7. `src/AcDream.Core/Physics/ServerControlledLocomotion.cs` — speed
+   bucket thresholds in `PlanFromVelocity`
+8. `docs/research/named-retail/acclient_2013_pseudo_c.txt` — grep by
+   the symbol names you discover from the TTD trace
+9. `references/ACE/Source/ACE.Server/Network/GameMessages/Messages/GameMessageUpdateMotion.cs`
+   — what ACE actually sends for each transition (ground truth for
+   wire content)
+
+---
+
+## Watchouts
+
+- **The named-retail leaf `MoveToStatePack::UnPack` returned 0 hits in
+  TTD.** Don't query it directly; start at `CPhysicsObj::unpack_movement`
+  or `MovementManager::unpack_movement` and walk down the call chain.
+  Release-build inlining + virtual dispatch hide leaf decoders from
+  TTD's static call counting (#44 lesson).
+- **Verify positioning IN-CLIENT before recording.** A trace where
+  retail #1 can't see retail #2 is wasted bytes. Walk both characters
+  to within visible range, confirm the actor is rendered on the
+  observer's screen, THEN attach TTD. The 30-min validation experiment
+  on 2026-05-06 wasted its main probe because of this — don't repeat.
+- **Outbound encoding quirk in our wire path** (CLAUDE.md "Outbound
+  motion wire format"): acdream sends `WalkForward (0x05) + HoldKey.Run`
+  for run, which ACE auto-upgrades to `RunForward (0x07)` when relaying
+  to remote observers. So the inbound parser sees `fwd=0x07` for
+  "remote is running." Don't get confused if outbound code shows 0x05
+  but inbound shows 0x07 — that's normal ACE behavior.
+- **TTD ring buffer with 512 MB max + ~60 sec recording** should fit
+  the whole scenario without rollover. If you record longer, consider
+  bumping to 1024 MB or chopping the scenario into multiple recordings.
+- **Don't kill the TTD process** with Stop-Process — it kills retail
+  too. Use Ctrl+C in the TTD window or `TTD.exe -stop <pid>` from
+  another elevated shell.
+
+---
+
+## Definition of done
+
+This investigation is done when:
+
+1. The TTD trace + queries identify the exact retail function(s) that
+   refine the cycle from UP-derived velocity. Function name(s) +
+   address(es) documented in `docs/research/2026-05-06-locomotion-cycle-transitions/findings.md`.
+2. The threshold/grace logic retail uses is documented (timing values,
+   conditions).
+3. `ApplyServerControlledVelocityCycle` un-gating shipped in acdream
+   with the corresponding test and visual verification on all 7 phases.
+4. ISSUES.md #39 closed with the commit SHA.
+5. Memory `project_retail_motion_outbound.md` (or a new note for the
+   inbound side) gets the durable lessons appended.
+
+---
+
+## Time estimate
+
+- Recording: ~30 min (setup, two retail clients, scenario execution)
+- Initial query passes + decomp cross-reference: ~1–2 hours
+- Implementation + iteration: ~2–4 hours
+- Visual verification + commit: ~30 min
+
+Total: half a day to a full day depending on how clean retail's path is.
+
+If after the first TTD pass the retail receive code looks fundamentally
+different from what `ApplyServerControlledVelocityCycle` is doing, stop
+and reassess — the fix shape may need to change. Don't try to ram a
+mismatched approach through.

src/AcDream.App/Rendering/GameWindow.cs

@@ -172,6 +172,38 @@ public sealed class GameWindow : IDisposable
     // Diagnostic: hide a specific humanoid part (>=10 parts) at render.
     private static readonly int s_hidePartIndex =
         int.TryParse(Environment.GetEnvironmentVariable("ACDREAM_HIDE_PART"), out var hp) ? hp : -1;
+
+    // Issue #47 — opt in to retail's close-detail GfxObj selection on
+    // humanoid setups. When enabled, every per-part GfxObj id (after
+    // server AnimPartChanges are applied) is replaced with Degrades[0]
+    // from its DIDDegrade table when present. See GfxObjDegradeResolver
+    // for the full retail-decomp citation. Off by default while the fix
+    // bakes; flip to default-on once we've confirmed no scenery/setup
+    // regressions.
+    private static readonly bool s_retailCloseDegrades =
+        string.Equals(Environment.GetEnvironmentVariable("ACDREAM_RETAIL_CLOSE_DEGRADES"), "1", StringComparison.Ordinal);
+
+    /// <summary>
+    /// Issue #47 humanoid-setup detector. Matches Aluvian Male
+    /// (<c>0x02000001</c>) and the 34-part heritage sibling setups
+    /// (Aluvian Female, Sho M/F, Gharu M/F, Viamont/Empyrean, etc.)
+    /// by structure rather than id list: a humanoid setup has exactly
+    /// 34 parts, and the trailing attachment slots (parts 17–33) are
+    /// the AC null-part sentinel <c>0x010001EC</c>. Non-humanoid
+    /// 34-part setups (rare) won't have the sentinel pattern.
+    /// </summary>
+    private static bool IsIssue47HumanoidSetup(DatReaderWriter.DBObjs.Setup setup)
+    {
+        if (setup.Parts.Count != 34) return false;
+        const uint NullPartGfx = 0x010001ECu;
+        int nullSlots = 0;
+        for (int i = 17; i < setup.Parts.Count; i++)
+            if ((uint)setup.Parts[i] == NullPartGfx) nullSlots++;
+        // At least half of slots 17–33 wired to the null sentinel — enough
+        // to distinguish humanoids from any future 34-part creature setup.
+        return nullSlots >= 8;
+    }
+
     private readonly HashSet<SkyPesKey> _activeSkyPes = new();
     private readonly HashSet<SkyPesKey> _missingSkyPes = new();
 
@@ -383,6 +415,19 @@ public sealed class GameWindow : IDisposable
         /// </summary>
         public float MaxSeqSpeedSinceLastUP;
 
+        /// <summary>
+        /// Seconds-since-epoch timestamp of the most recent UpdateMotion (UM)
+        /// for this remote. Used by the player-remote velocity-fallback cycle
+        /// refinement to skip refinement while a fresh UM is authoritative —
+        /// retail's outbound MoveToState gives us direction-explicit cycles
+        /// on direction-key changes (W press, W release, W↔S flip), and we
+        /// only want UP-derived velocity to refine the speed bucket within
+        /// a direction when no UM has arrived recently. Defaults to 0
+        /// (epoch) so the first UP after spawn is allowed to refine
+        /// immediately if velocity already differs from the spawn cycle.
+        /// </summary>
+        public double LastUMTime;
+
         public RemoteMotion()
         {
             Body = new AcDream.Core.Physics.PhysicsBody
@@ -669,6 +714,13 @@ public sealed class GameWindow : IDisposable
     /// </summary>
     private readonly Dictionary<uint, AcDream.Core.World.WorldEntity> _entitiesByServerGuid = new();
     private readonly Dictionary<uint, LiveEntityInfo> _liveEntityInfoByGuid = new();
+    /// <summary>
+    /// Latest <see cref="AcDream.Core.Net.WorldSession.EntitySpawn"/> for each
+    /// guid. Captured at the end of <see cref="OnLiveEntitySpawnedLocked"/> so
+    /// <see cref="OnLiveAppearanceUpdated"/> can reuse the position/setup/motion
+    /// fields when a 0xF625 ObjDescEvent arrives carrying only updated visuals.
+    /// </summary>
+    private readonly Dictionary<uint, AcDream.Core.Net.WorldSession.EntitySpawn> _lastSpawnByGuid = new();
     private uint? _selectedTargetGuid;
     private readonly record struct LiveEntityInfo(
         string? Name,
@@ -1463,6 +1515,7 @@ public sealed class GameWindow : IDisposable
             _liveSession.PositionUpdated += OnLivePositionUpdated;
             _liveSession.VectorUpdated += OnLiveVectorUpdated;
             _liveSession.TeleportStarted += OnTeleportStarted;
+            _liveSession.AppearanceUpdated += OnLiveAppearanceUpdated;
 
             // Phase 6c — PlayScript (0xF754) arrives from the server as
             // a (guid, scriptId) pair. Resolve the guid's current world
@@ -1975,7 +2028,34 @@ public sealed class GameWindow : IDisposable
             && setup.Parts.Count >= 10;
         if (dumpClothing)
         {
-            Console.WriteLine($"\n=== DUMP_CLOTHING: guid=0x{spawn.Guid:X8} name='{spawn.Name}' setup=0x{setup.Id:X8} APC={animPartChanges.Count} ===");
+            Console.WriteLine($"\n=== DUMP_CLOTHING: guid=0x{spawn.Guid:X8} name='{spawn.Name}' setup=0x{setup.Id:X8} setup.Parts.Count={setup.Parts.Count} flatten.Count={flat.Count} APC={animPartChanges.Count} ===");
+            // Dump the Setup's ParentIndex + DefaultScale arrays to verify hierarchy.
+            var parentStr = string.Join(",", setup.ParentIndex.Take(Math.Min(34, setup.ParentIndex.Count)).Select(p => p == 0xFFFFFFFFu ? "-1" : p.ToString()));
+            Console.WriteLine($"  ParentIndex[{setup.ParentIndex.Count}]: {parentStr}");
+            var scaleStr = string.Join(",", setup.DefaultScale.Take(Math.Min(34, setup.DefaultScale.Count)).Select(s => $"({s.X:F2},{s.Y:F2},{s.Z:F2})"));
+            Console.WriteLine($"  DefaultScale[{setup.DefaultScale.Count}]: {scaleStr}");
+            // Dump the resolved idle frame's per-part Origin + Orientation.
+            // If retail composes parent_world * animation_local but acdream
+            // treats animation_local as world-relative, we'd see specific
+            // patterns of non-zero per-part origins/rotations that should
+            // be parent-relative. For setups whose idle has all parts at
+            // (0,0,0)/identity, parent walking would be a no-op (which
+            // matches my earlier "no change" experiment if that was the
+            // human-idle case) — diagnostic confirms.
+            if (idleFrame is not null)
+            {
+                Console.WriteLine($"  IdleFrame.Frames[{idleFrame.Frames.Count}]:");
+                int dumpCount = Math.Min(idleFrame.Frames.Count, 17);  // first 17 (real body parts, not the 17-33 placeholders)
+                for (int fi = 0; fi < dumpCount; fi++)
+                {
+                    var f = idleFrame.Frames[fi];
+                    Console.WriteLine($"    [{fi:D2}] Origin=({f.Origin.X:F3},{f.Origin.Y:F3},{f.Origin.Z:F3}) Orient=(W={f.Orientation.W:F3} X={f.Orientation.X:F3} Y={f.Orientation.Y:F3} Z={f.Orientation.Z:F3})");
+                }
+            }
+            else
+            {
+                Console.WriteLine($"  IdleFrame: NULL");
+            }
             foreach (var c in animPartChanges)
                 Console.WriteLine($"  APC part={c.PartIndex:D2} -> gfx=0x{c.NewModelId:X8}");
 
@@ -2028,6 +2108,41 @@ public sealed class GameWindow : IDisposable
             }
         }
 
+        // Issue #47 — retail's close/player rendering path resolves each
+        // part's base GfxObj through its DIDDegrade table to the close-
+        // detail mesh in slot 0. Without this, humanoid arms/torso draw
+        // the LOW-detail base GfxObj (e.g. 0x01000055, 14 verts / 17
+        // polys) instead of the close mesh (0x01001795, 32 verts / 60
+        // polys), losing all bicep/shoulder/back geometry. See
+        // <see cref="GfxObjDegradeResolver"/> for the named-retail
+        // citation (CPhysicsPart::LoadGfxObjArray at 0x0050DCF0,
+        // ::UpdateViewerDistance at 0x0050E030, ::Draw at 0x0050D7A0).
+        //
+        // Order matters: the swap happens AFTER AnimPartChanges have
+        // installed the server's body/clothing/head ids, BEFORE texture
+        // changes resolve (which match against the resolved mesh's
+        // surfaces) and BEFORE the GfxObjMesh.Build / texture upload
+        // path consumes the part list.
+        if (s_retailCloseDegrades && IsIssue47HumanoidSetup(setup))
+        {
+            for (int partIdx = 0; partIdx < parts.Count; partIdx++)
+            {
+                var part = parts[partIdx];
+                if (!AcDream.Core.Meshing.GfxObjDegradeResolver.TryResolveCloseGfxObj(
+                        _dats, part.GfxObjId,
+                        out uint resolvedId, out _))
+                    continue;
+                if (resolvedId == part.GfxObjId)
+                    continue;
+
+                parts[partIdx] = new AcDream.Core.World.MeshRef(
+                    resolvedId, part.PartTransform);
+
+                if (dumpClothing)
+                    Console.WriteLine($"  DEGRADE part={partIdx:D2} gfx=0x{part.GfxObjId:X8} -> close=0x{resolvedId:X8}");
+            }
+        }
+
         // Build per-part texture overrides. The server sends TextureChanges as
         // (partIdx, oldSurfaceTextureId, newSurfaceTextureId) where both ids
         // are in the SurfaceTexture (0x05) range. Our sub-meshes are keyed
@@ -2145,14 +2260,27 @@ public sealed class GameWindow : IDisposable
         var scaleMat = System.Numerics.Matrix4x4.CreateScale(scale);
 
         var meshRefs = new List<AcDream.Core.World.MeshRef>();
+        int dumpClothingTotalTris = 0;
         for (int partIdx = 0; partIdx < parts.Count; partIdx++)
         {
             var mr = parts[partIdx];
             var gfx = _dats.Get<DatReaderWriter.DBObjs.GfxObj>(mr.GfxObjId);
-            if (gfx is null) continue;
+            if (gfx is null)
+            {
+                if (dumpClothing)
+                    Console.WriteLine($"  EMIT part={partIdx:D2} gfx=0x{mr.GfxObjId:X8} GFXOBJ_DAT_MISSING -> 0 tris");
+                continue;
+            }
             _physicsDataCache.CacheGfxObj(mr.GfxObjId, gfx);
             var subMeshes = AcDream.Core.Meshing.GfxObjMesh.Build(gfx, _dats);
             _staticMesh.EnsureUploaded(mr.GfxObjId, subMeshes);
+            if (dumpClothing)
+            {
+                int tris = 0; int subs = 0;
+                foreach (var sm in subMeshes) { tris += sm.Indices.Length / 3; subs++; }
+                dumpClothingTotalTris += tris;
+                Console.WriteLine($"  EMIT part={partIdx:D2} gfx=0x{mr.GfxObjId:X8} subMeshes={subs} tris={tris}");
+            }
 
             IReadOnlyDictionary<uint, uint>? surfaceOverrides = null;
             if (resolvedOverridesByPart is not null && resolvedOverridesByPart.TryGetValue(partIdx, out var partOverrides))
@@ -2181,6 +2309,8 @@ public sealed class GameWindow : IDisposable
                               $"(guid=0x{spawn.Guid:X8})");
             return;
         }
+        if (dumpClothing)
+            Console.WriteLine($"  TOTAL tris={dumpClothingTotalTris} meshRefs={meshRefs.Count} (parts.Count={parts.Count})");
 
         // Build optional per-entity palette override from the server's base
         // palette + subpalette overlays. The renderer applies these to
@@ -2228,6 +2358,10 @@ public sealed class GameWindow : IDisposable
         // UpdateMotion / UpdatePosition events can reseat this entity by guid.
         _entitiesByServerGuid[spawn.Guid] = entity;
 
+        // Cache the spawn so OnLiveAppearanceUpdated can replay it with new
+        // appearance fields when a later 0xF625 ObjDescEvent arrives.
+        _lastSpawnByGuid[spawn.Guid] = spawn;
+
         // Commit B 2026-04-29 — live-entity collision registration. The
         // local player is the simulator (its PhysicsBody is the source of
         // truth for our own movement); only remotes register as targets.
@@ -2457,6 +2591,40 @@ public sealed class GameWindow : IDisposable
         }
     }
 
+    /// <summary>
+    /// Server broadcast a <c>0xF625 ObjDescEvent</c> — a creature/player's
+    /// appearance changed (equip / unequip / tailoring / recipe result /
+    /// character option toggle). The wire payload only carries the new
+    /// ModelData (palette + texture + animpart changes), not position or
+    /// motion, so we splice it onto the cached spawn and replay through
+    /// <see cref="OnLiveEntitySpawned"/>. The dedup at the start of
+    /// <see cref="OnLiveEntitySpawnedLocked"/> tears down the previous
+    /// rendering state (GpuWorldState entry, animated entity, collision
+    /// registration) before rebuilding.
+    /// </summary>
+    private void OnLiveAppearanceUpdated(AcDream.Core.Net.Messages.ObjDescEvent.Parsed update)
+    {
+        if (!_lastSpawnByGuid.TryGetValue(update.Guid, out var oldSpawn))
+        {
+            // Server can broadcast ObjDescEvent before we've seen a
+            // CreateObject for this guid (race on landblock entry, or
+            // if the entity is in a state we couldn't render). Drop —
+            // when CreateObject lands, ACE includes the same ModelData
+            // body inside it, so the appearance won't be lost.
+            return;
+        }
+
+        var md = update.ModelData;
+        var newSpawn = oldSpawn with
+        {
+            AnimPartChanges = md.AnimPartChanges,
+            TextureChanges = md.TextureChanges,
+            SubPalettes = md.SubPalettes,
+            BasePaletteId = md.BasePaletteId,
+        };
+        OnLiveEntitySpawned(newSpawn);
+    }
+
     /// <summary>
     /// Commit B 2026-04-29 — register a live (server-spawned) entity into
     /// the <see cref="ShadowObjectRegistry"/> as a single collision body.
@@ -2567,6 +2735,7 @@ public sealed class GameWindow : IDisposable
         _remoteLastMove.Remove(serverGuid);
         _liveEntityInfoByGuid.Remove(serverGuid);
         _entitiesByServerGuid.Remove(serverGuid);
+        _lastSpawnByGuid.Remove(serverGuid);
         if (_selectedTargetGuid == serverGuid)
             _selectedTargetGuid = null;
 
@@ -2590,6 +2759,19 @@ public sealed class GameWindow : IDisposable
         if (!_entitiesByServerGuid.TryGetValue(update.Guid, out var entity)) return;
         if (!_animatedEntities.TryGetValue(entity.Id, out var ae)) return;
 
+        // #39 (2026-05-06): stamp the per-remote LastUMTime so the
+        // UP-velocity fallback path in ApplyServerControlledVelocityCycle
+        // can skip refinement while a UM is fresh. UMs are authoritative
+        // for direction-key changes (W press / release / W↔S flip);
+        // velocity refinement only helps for HoldKey-only changes (Shift
+        // toggle while a direction key is held — retail does NOT broadcast
+        // a fresh MoveToState in that case).
+        if (_remoteDeadReckon.TryGetValue(update.Guid, out var rmStateForUm))
+        {
+            rmStateForUm.LastUMTime =
+                (System.DateTime.UtcNow - System.DateTime.UnixEpoch).TotalSeconds;
+        }
+
         // Re-resolve using the new stance/command. Keep the setup and
         // motion-table we already know about — the server's motion
         // updates override state within the same table, not swap tables.
@@ -3317,13 +3499,41 @@ public sealed class GameWindow : IDisposable
         return low is 0x05 or 0x06 or 0x07 or 0x0F or 0x10;
     }
 
+    /// <summary>
+    /// Grace window in seconds after a UM arrives during which UP-derived
+    /// velocity refinement is suppressed for a player remote. UMs are
+    /// authoritative; the velocity fallback only fills the gap when retail
+    /// does not send a fresh MoveToState (Shift toggle while direction key
+    /// held). 200 ms covers the worst-case UM/UP race — UMs arrive on
+    /// direction-key events and UPs at 5–10 Hz, so the first UP after a
+    /// fresh UM lands ~100–200 ms behind. Tightened from 500 ms (commit
+    /// 8fa04af original) per user observation that the Shift-toggle
+    /// transition was visibly slower than retail; with 0.2 s the worst-case
+    /// added latency is just the UP cadence below it.
+    /// </summary>
+    private const double UmGraceSeconds = 0.2;
+
+    /// <summary>
+    /// Speed (m/s) above which a player-remote currently in WalkForward
+    /// is promoted to RunForward by velocity refinement. Tuned to player
+    /// speeds: walk ≈ 3.12 m/s (WalkAnimSpeed × 1.0), run ≈ 8–12 m/s
+    /// (RunAnimSpeed × runRate ≈ 4.0 × 2.0–3.0). Hysteresis with
+    /// <see cref="PlayerRunDemoteSpeed"/> avoids thrashing at the boundary.
+    /// </summary>
+    private const float PlayerRunPromoteSpeed = 5.5f;
+
+    /// <summary>
+    /// Speed (m/s) below which a player-remote currently in RunForward
+    /// is demoted to WalkForward by velocity refinement.
+    /// </summary>
+    private const float PlayerRunDemoteSpeed = 4.5f;
+
     private void ApplyServerControlledVelocityCycle(
         uint serverGuid,
         AnimatedEntity ae,
         RemoteMotion rm,
         System.Numerics.Vector3 velocity)
     {
-        if (IsPlayerGuid(serverGuid)) return;
         if (rm.Airborne) return;
         if (ae.Sequencer is null) return;
         // MoveTo packets already seeded the retail speed/runRate cycle.
@@ -3332,6 +3542,32 @@ public sealed class GameWindow : IDisposable
         // velocity-estimated animation.
         if (rm.ServerMoveToActive) return;
 
+        if (IsPlayerGuid(serverGuid))
+        {
+            // #39 (2026-05-06): player-remote forward-direction speed-bucket
+            // refinement. The bug case: actor toggles Shift while holding W
+            // (or releases Shift). Retail's outbound apparently does NOT
+            // broadcast a fresh MoveToState for HoldKey-only changes
+            // (verified via static analysis of CommandInterpreter::SendMovementEvent
+            // call sites; needs cdb confirmation). ACE has nothing to
+            // broadcast → no UM arrives at the observer → cycle stays at
+            // whichever direction-bucket was last set. Velocity DOES change
+            // (UP carries new pace), so this code path uses UP-derived
+            // velocity to refine the speed bucket within the same direction.
+            //
+            // Conservative scope:
+            //  - Forward direction only (low byte 0x05 or 0x07). Sidestep
+            //    and backward HoldKey toggles are deferred until the TTD
+            //    trace described in
+            //    docs/research/2026-05-06-locomotion-cycle-transitions/
+            //    confirms retail's exact algorithm.
+            //  - Hysteresis (4.5 m/s demote / 5.5 m/s promote) prevents
+            //    thrashing at the boundary.
+            //  - 500 ms UM grace window — a fresh UM is always authoritative.
+            ApplyPlayerLocomotionRefinement(serverGuid, ae, rm, velocity);
+            return;
+        }
+
         var plan = AcDream.Core.Physics.ServerControlledLocomotion
             .PlanFromVelocity(velocity);
         uint currentMotion = ae.Sequencer.CurrentMotion;
@@ -3344,10 +3580,7 @@ public sealed class GameWindow : IDisposable
 
         // D2 (Commit A 2026-05-03): UPCYCLE diag — proves whether
         // ApplyServerControlledVelocityCycle is racing UpdateMotion-driven
-        // SetCycle for player-driven remotes. If this fires every ~100-200ms
-        // during a Walk→Run press with `motion` flipping between buckets,
-        // H2 (UP-vs-UM race) is confirmed. UPs (5-10 Hz) would then
-        // perpetually overwrite the cycle the UM just set.
+        // SetCycle for non-player remotes (NPCs / monsters).
         if (System.Environment.GetEnvironmentVariable("ACDREAM_REMOTE_VEL_DIAG") == "1")
         {
             System.Console.WriteLine(
@@ -3361,6 +3594,162 @@ public sealed class GameWindow : IDisposable
         ae.Sequencer.SetCycle(style, plan.Motion, plan.SpeedMod);
     }
 
+    private void ApplyPlayerLocomotionRefinement(
+        uint serverGuid,
+        AnimatedEntity ae,
+        RemoteMotion rm,
+        System.Numerics.Vector3 velocity)
+    {
+        // UM grace: a fresh UM is authoritative.
+        double nowSec = (System.DateTime.UtcNow - System.DateTime.UnixEpoch).TotalSeconds;
+        double sinceUm = nowSec - rm.LastUMTime;
+        if (sinceUm < UmGraceSeconds) return;
+
+        uint currentMotion = ae.Sequencer!.CurrentMotion;
+        uint lowByte = currentMotion & 0xFFu;
+        float currentSign = MathF.Sign(ae.Sequencer.CurrentSpeedMod);
+        if (currentSign == 0f) currentSign = 1f;
+
+        // Recognised locomotion directions:
+        //   0x05 (WalkForward) — also encodes WalkBackward via negative speed
+        //                        (ACE convention: SidestepCommand= cancel, ForwardCommand=
+        //                        WalkForward, ForwardSpeed *= -0.65)
+        //   0x07 (RunForward)
+        //   0x0F (SideStepRight)
+        //   0x10 (SideStepLeft)
+        // Other motions (Ready, Turn, emotes, attacks) are left to UM-driven SetCycle.
+        const uint LowWalkForward     = 0x05u;
+        const uint LowRunForward      = 0x07u;
+        const uint LowSideStepRight   = 0x0Fu;
+        const uint LowSideStepLeft    = 0x10u;
+        bool isForwardClass = lowByte == LowWalkForward || lowByte == LowRunForward;
+        bool isSidestep = lowByte == LowSideStepRight || lowByte == LowSideStepLeft;
+        if (!isForwardClass && !isSidestep) return;
+
+        float horizSpeed = MathF.Sqrt(velocity.X * velocity.X + velocity.Y * velocity.Y);
+
+        // Hysteresis: stay in current bucket unless we cross the appropriate
+        // threshold. Below StopSpeed → don't refine (let UM Ready stop signal
+        // handle the stop transition; we don't want UP momentary 0-velocity
+        // to drop the cycle to Ready while the actor is mid-stride).
+        if (horizSpeed < AcDream.Core.Physics.ServerControlledLocomotion.StopSpeed)
+            return;
+
+        uint targetMotion;
+        float speedMod;
+
+        if (isSidestep)
+        {
+            // Sidestep: motion ID stays the same (SideStepLeft / SideStepRight).
+            // Retail's wire encoding for sidestep speed buckets uses the same
+            // motion ID with different SidestepSpeed (slow ≈ 1.25 multiplier,
+            // fast ≈ 3.0 clamp per ACE MovementData.cs:124-131). On Shift
+            // toggle while a strafe key is held, retail does NOT broadcast a
+            // fresh MoveToState (same wire-silence rule as the forward case),
+            // so observer-side cycle refinement must come from UP-derived
+            // velocity here. Preserve the sign — SideStepLeft is sometimes
+            // emitted with negative speedMod by the adjust_motion path.
+            //
+            // Magnitude: horizSpeed / SidestepAnimSpeed maps the observed
+            // speed back to a SideStepSpeed the sequencer can apply as a
+            // framerate multiplier. Retail's get_state_velocity for
+            // sidestep cycles is `velocity.X = SidestepAnimSpeed *
+            // SideStepSpeed` (MotionInterpreter.cs:592 — constant 1.25
+            // m/s). Dividing by WalkAnimSpeed (3.12) here was wrong by
+            // 2.5× and made slow strafe play visibly slower than retail.
+            float sideMag = horizSpeed / AcDream.Core.Physics.MotionInterpreter.SidestepAnimSpeed;
+            sideMag = MathF.Min(MathF.Max(
+                sideMag,
+                AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            targetMotion = currentMotion;
+            speedMod = sideMag * currentSign;
+        }
+        else if (currentSign < 0f)
+        {
+            // BACKWARD walk: ACE encodes WalkBackward as `WalkForward` motion
+            // with NEGATIVE speedMod (MovementData.cs:115 `interpState.ForwardSpeed *= -0.65f`).
+            // No "RunBackward" motion exists — Shift toggle on backward
+            // changes only the magnitude of speedMod (slow back ≈ -0.65,
+            // fast back ≈ -1.91 = -runRate × 0.65). Keep WalkForward motion,
+            // refine magnitude, preserve negative sign.
+            //
+            // Without this branch (the original fix #1), backward refinement
+            // computed a positive speedMod from horizSpeed and overwrote the
+            // negative sign, making the legs play forward-walk while the body
+            // continued moving backward (the rubber-banding the user reported).
+            float backMag = horizSpeed / AcDream.Core.Physics.MotionInterpreter.WalkAnimSpeed;
+            backMag = MathF.Min(MathF.Max(
+                backMag,
+                AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            targetMotion = AcDream.Core.Physics.MotionCommand.WalkForward;
+            speedMod = -backMag;
+        }
+        else if (lowByte == LowRunForward)
+        {
+            if (horizSpeed < PlayerRunDemoteSpeed)
+            {
+                targetMotion = AcDream.Core.Physics.MotionCommand.WalkForward;
+                speedMod = MathF.Min(MathF.Max(
+                    horizSpeed / AcDream.Core.Physics.MotionInterpreter.WalkAnimSpeed,
+                    AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                    AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            }
+            else
+            {
+                targetMotion = AcDream.Core.Physics.MotionCommand.RunForward;
+                speedMod = MathF.Min(MathF.Max(
+                    horizSpeed / AcDream.Core.Physics.MotionInterpreter.RunAnimSpeed,
+                    AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                    AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            }
+        }
+        else
+        {
+            // currently WalkForward (0x05) with positive speedMod = walking forward.
+            if (horizSpeed > PlayerRunPromoteSpeed)
+            {
+                targetMotion = AcDream.Core.Physics.MotionCommand.RunForward;
+                speedMod = MathF.Min(MathF.Max(
+                    horizSpeed / AcDream.Core.Physics.MotionInterpreter.RunAnimSpeed,
+                    AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                    AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            }
+            else
+            {
+                targetMotion = AcDream.Core.Physics.MotionCommand.WalkForward;
+                speedMod = MathF.Min(MathF.Max(
+                    horizSpeed / AcDream.Core.Physics.MotionInterpreter.WalkAnimSpeed,
+                    AcDream.Core.Physics.ServerControlledLocomotion.MinSpeedMod),
+                    AcDream.Core.Physics.ServerControlledLocomotion.MaxSpeedMod);
+            }
+        }
+
+        // Skip the SetCycle if neither motion nor speedMod changed
+        // meaningfully — avoids replaying transition links every UP.
+        bool motionChanged = currentMotion != targetMotion
+            && (currentMotion & 0xFFu) != (targetMotion & 0xFFu);
+        bool speedChanged = MathF.Abs(ae.Sequencer.CurrentSpeedMod - speedMod) > 0.05f;
+        if (!motionChanged && !speedChanged)
+            return;
+
+        if (System.Environment.GetEnvironmentVariable("ACDREAM_REMOTE_VEL_DIAG") == "1")
+        {
+            System.Console.WriteLine(
+                $"[UPCYCLE_PLAYER] guid={serverGuid:X8} "
+                + $"|v|={horizSpeed:F2} cur=0x{currentMotion:X8} "
+                + $"-> motion=0x{targetMotion:X8} speedMod={speedMod:F2} "
+                + $"sinceUM={sinceUm:F2}s "
+                + $"motionChg={motionChanged} speedChg={speedChanged}");
+        }
+
+        uint style = ae.Sequencer.CurrentStyle != 0
+            ? ae.Sequencer.CurrentStyle
+            : 0x8000003Du;
+        ae.Sequencer.SetCycle(style, targetMotion, speedMod);
+    }
+
     private void OnLivePositionUpdated(AcDream.Core.Net.WorldSession.EntityPositionUpdate update)
     {
         // Phase A.1: track the most recently updated entity's landblock so the
@@ -3381,6 +3770,13 @@ public sealed class GameWindow : IDisposable
         var rot = new System.Numerics.Quaternion(p.RotationX, p.RotationY, p.RotationZ, p.RotationW);
         DumpMovementTruthServerEcho(update, worldPos);
 
+        // Keep the cached spawn's Position in sync with server truth so a
+        // later ObjDescEvent (which only carries new appearance, not new
+        // position) re-applies at the entity's CURRENT location instead of
+        // popping back to its login spot. See OnLiveAppearanceUpdated.
+        if (_lastSpawnByGuid.TryGetValue(update.Guid, out var cached))
+            _lastSpawnByGuid[update.Guid] = cached with { Position = update.Position };
+
         // Capture the pre-update render position for the soft-snap residual
         // calculation below. Assign entity.Position to the server truth up
         // front; if we then compute a snap residual, we restore the rendered
@@ -3574,6 +3970,51 @@ public sealed class GameWindow : IDisposable
                         isMovingTo: false,
                         currentBodyPosition: rmState.Body.Position);
                 }
+                // #39 fix-3 (2026-05-06): velocity-fallback cycle refinement
+                // for player remotes. Wire-level evidence (`launch-39-diag2.log`):
+                // when retail's actor toggles Shift while a direction key
+                // is held, retail's outbound MoveToState logic does NOT
+                // emit a fresh packet (only Ready ↔ Run UMs visible in
+                // `[FWD_WIRE]`, despite a clear walk-pace ↔ run-pace
+                // velocity transition in `[VEL_DIAG]`). ACE has nothing
+                // to broadcast → no UM arrives at the observer → cycle
+                // sticks at whatever the last UM set. Compute the
+                // synth-velocity here in the player-remote path AND
+                // call into ApplyServerControlledVelocityCycle, which
+                // routes through the direction-preserving + UM-grace
+                // ApplyPlayerLocomotionRefinement helper (added in
+                // commit 8fa04af).
+                //
+                // The legacy non-player block below (3759+) covers NPCs
+                // and is gated `!IsPlayerGuid`; this block fills the
+                // matching gap for players.
+                if (rmState.PrevServerPosTime > 0.0)
+                {
+                    double nowSecVel = rmState.LastServerPosTime;
+                    double dtPos = nowSecVel - rmState.PrevServerPosTime;
+                    if (dtPos > 0.001)
+                    {
+                        var synthVel = (worldPos - rmState.PrevServerPos) / (float)dtPos;
+                        rmState.ServerVelocity = synthVel;
+                        rmState.HasServerVelocity = true;
+
+                        if (_animatedEntities.TryGetValue(entity.Id, out var aeForVel)
+                            && aeForVel.Sequencer is not null)
+                        {
+                            if (System.Environment.GetEnvironmentVariable("ACDREAM_REMOTE_VEL_DIAG") == "1")
+                            {
+                                System.Console.WriteLine(
+                                    $"[UPCYCLE_SRC] guid={update.Guid:X8} src=synth-player");
+                            }
+                            ApplyServerControlledVelocityCycle(
+                                update.Guid,
+                                aeForVel,
+                                rmState,
+                                synthVel);
+                        }
+                    }
+                }
+
                 // Sync the visible entity to the body — overrides the unconditional
                 // entity.Position = worldPos snap at the top of this function.
                 // For the far-snap branch this is a no-op (body == worldPos); for
@@ -3685,15 +4126,20 @@ public sealed class GameWindow : IDisposable
                 rmState.Body.Velocity = rmState.ServerVelocity;
             }
 
-            if (!IsPlayerGuid(update.Guid)
-                && rmState.HasServerVelocity
+            if (rmState.HasServerVelocity
                 && _animatedEntities.TryGetValue(entity.Id, out var aeForVelocity))
             {
+                // #39 (2026-05-06): un-gated for player remotes — the
+                // function itself routes player remotes into the dedicated
+                // ApplyPlayerLocomotionRefinement path (forward-direction
+                // speed bucket only, with UM grace + hysteresis). Non-player
+                // remotes use the existing PlanFromVelocity path.
+                //
                 // D2 (Commit A 2026-05-03): tag whether the velocity feeding
                 // ApplyServerControlledVelocityCycle is wire-explicit (rare for
                 // player remotes — ACE almost never sets HasVelocity on player
                 // UPs) or synthesized from position deltas (the common case).
-                // Pairs with the [UPCYCLE] line printed inside the call.
+                // Pairs with the [UPCYCLE]/[UPCYCLE_PLAYER] line printed inside.
                 if (System.Environment.GetEnvironmentVariable("ACDREAM_REMOTE_VEL_DIAG") == "1")
                 {
                     string velSrc = update.Velocity is null ? "synth" : "wire";
@@ -6629,6 +7075,23 @@ public sealed class GameWindow : IDisposable
                         System.Console.WriteLine(
                             $"[SEQSTATE] guid={serverGuid:X8} CurrentMotion=0x{ae.Sequencer.CurrentMotion:X8} "
                             + $"CurrentSpeedMod={ae.Sequencer.CurrentSpeedMod:F3}");
+
+                        // #39 fix-3 evidence (2026-05-06): CURRNODE proves
+                        // whether _currNode is actually on the cycle (anim
+                        // ref hash matches FirstCyclic) or stuck somewhere
+                        // else. SCFULL captures _currNode==_firstCyclic only
+                        // at SetCycle return; this captures it per render
+                        // tick so we can see if something resets it later.
+                        var d = ae.Sequencer.CurrentNodeDiag;
+                        int firstHash = ae.Sequencer.FirstCyclicAnimRefHash;
+                        System.Console.WriteLine(
+                            $"[CURRNODE] guid={serverGuid:X8} "
+                            + $"animRef=0x{d.AnimRefHash:X8} firstCyclicAnimRef=0x{firstHash:X8} "
+                            + $"isOnCyclic={d.AnimRefHash == firstHash && firstHash != 0} "
+                            + $"isLooping={d.IsLooping} fr={d.Framerate:F2} "
+                            + $"frame=[{d.StartFrame}..{d.EndFrame}] "
+                            + $"pos={d.FramePosition:F2} qCount={d.QueueCount}");
+
                         rmDiag.LastSeqStateLogTime = nowSec;
                     }
                 }
@@ -6944,6 +7407,27 @@ public sealed class GameWindow : IDisposable
             // For everything else (Walk → Run, Run → Ready, etc.) we
             // keep the link so transitions stay smooth.
             bool skipLink = animCommand == AcDream.Core.Physics.MotionCommand.Falling;
+
+            // #45 (2026-05-06): scale sidestep speedMod to match ACE's
+            // wire formula. PlayerMovementController hands us a raw
+            // localAnimSpeed (1.0 slow / runRate fast), but ACE's
+            // BroadcastMovement converts SidestepSpeed via
+            //   `speed × 3.12 / 1.25 × 0.5`
+            // (Network/Motion/MovementData.cs:124-131). Without the
+            // matching multiplier here, the local sidestep cycle plays
+            // at speedMod = 1.0 while the observer-side cycle plays at
+            // ~1.248 — local strafe visibly slower than retail (user
+            // report at #45 close-out of #39).
+            // Factor = WalkAnimSpeed / SidestepAnimSpeed × 0.5
+            //        = 3.12 / 1.25 × 0.5 = 1.248.
+            uint cmdLow = animCommand & 0xFFu;
+            if (cmdLow == 0x0Fu /* SideStepRight */ || cmdLow == 0x10u /* SideStepLeft */)
+            {
+                animSpeed *= AcDream.Core.Physics.MotionInterpreter.WalkAnimSpeed
+                           / AcDream.Core.Physics.MotionInterpreter.SidestepAnimSpeed
+                           * 0.5f;
+            }
+
             ae.Sequencer.SetCycle(fullStyle, animCommand, animSpeed * animScale,
                 skipTransitionLink: skipLink);
         }

src/AcDream.Core.Net/Messages/CreateObject.cs

@@ -269,6 +269,18 @@ public static class CreateObject
     /// </summary>
     public readonly record struct AnimPartChange(byte PartIndex, uint NewModelId);
 
+    /// <summary>
+    /// The ModelData block — palette/texture/animpart changes — that lives
+    /// inside both CreateObject (initial spawn) and ObjDescEvent (0xF625
+    /// appearance update). Factored out so both sites parse the same wire
+    /// shape with one implementation.
+    /// </summary>
+    public readonly record struct ModelData(
+        uint? BasePaletteId,
+        IReadOnlyList<SubPaletteSwap> SubPalettes,
+        IReadOnlyList<TextureChange> TextureChanges,
+        IReadOnlyList<AnimPartChange> AnimPartChanges);
+
     /// <summary>
     /// Parse a reassembled CreateObject body. <paramref name="body"/> must
     /// start with the 4-byte opcode. Returns <c>null</c> if the body is
@@ -310,64 +322,11 @@ public static class CreateObject
 
             uint guid = ReadU32(body, ref pos);
 
-            // --- ModelData ---
-            // Header: byte 0x11 marker, byte subPalettes, byte textureChanges, byte animPartChanges
-            if (body.Length - pos < 4) return null;
-            byte _marker = body[pos]; pos += 1;
-            byte subPaletteCount = body[pos]; pos += 1;
-            byte textureChangeCount = body[pos]; pos += 1;
-            byte animPartChangeCount = body[pos]; pos += 1;
-
-            uint? basePaletteId = null;
-            if (subPaletteCount > 0)
-                basePaletteId = ReadPackedDwordOfKnownType(body, ref pos, PaletteTypePrefix);
-
-            var subPalettes = subPaletteCount == 0
-                ? (IReadOnlyList<SubPaletteSwap>)Array.Empty<SubPaletteSwap>()
-                : new SubPaletteSwap[subPaletteCount];
-            for (int i = 0; i < subPaletteCount; i++)
-            {
-                uint subPalId = ReadPackedDwordOfKnownType(body, ref pos, PaletteTypePrefix);
-                if (body.Length - pos < 2) return null;
-                byte offset = body[pos]; pos += 1;
-                byte length = body[pos]; pos += 1;
-                ((SubPaletteSwap[])subPalettes)[i] = new SubPaletteSwap(subPalId, offset, length);
-            }
-
-            var textureChanges = textureChangeCount == 0
-                ? (IReadOnlyList<TextureChange>)Array.Empty<TextureChange>()
-                : new TextureChange[textureChangeCount];
-            for (int i = 0; i < textureChangeCount; i++)
-            {
-                if (body.Length - pos < 1) return null;
-                byte partIndex = body[pos]; pos += 1;
-                uint oldTex = ReadPackedDwordOfKnownType(body, ref pos, SurfaceTextureTypePrefix);
-                uint newTex = ReadPackedDwordOfKnownType(body, ref pos, SurfaceTextureTypePrefix);
-                ((TextureChange[])textureChanges)[i] = new TextureChange(partIndex, oldTex, newTex);
-            }
-
-            // Extract AnimPartChanges — the server uses these to replace
-            // base Setup parts with armored/statue/whatever-specific meshes.
-            // Without decoding these, characters render "naked" and custom
-            // weenies render as whatever their base Setup looks like.
-            //
-            // NOTE: ACE writes the NewModelId through WritePackedDwordOfKnownType
-            // with knownType=0x01000000 (GfxObj type prefix). That writer STRIPS
-            // the high-byte type if present before writing the PackedDword. We
-            // have to OR it back on read or our GfxObj dat lookup will fail
-            // (silently, producing no mesh refs — hence the Phase 4.7h regression).
-            var animParts = animPartChangeCount == 0
-                ? (IReadOnlyList<AnimPartChange>)Array.Empty<AnimPartChange>()
-                : new AnimPartChange[animPartChangeCount];
-            for (int i = 0; i < animPartChangeCount; i++)
-            {
-                if (body.Length - pos < 1) return null;
-                byte partIndex = body[pos]; pos += 1;
-                uint newModelId = ReadPackedDwordOfKnownType(body, ref pos, GfxObjTypePrefix);
-                ((AnimPartChange[])animParts)[i] = new AnimPartChange(partIndex, newModelId);
-            }
-
-            AlignTo4(ref pos);
+            var modelData = ReadModelData(body, ref pos);
+            uint? basePaletteId = modelData.BasePaletteId;
+            var subPalettes = modelData.SubPalettes;
+            var textureChanges = modelData.TextureChanges;
+            var animParts = modelData.AnimPartChanges;
 
             // --- PhysicsData ---
             if (body.Length - pos < 8) return null;
@@ -559,6 +518,80 @@ public static class CreateObject
         }
     }
 
+    /// <summary>
+    /// Read the ModelData block — palette swaps + texture overrides +
+    /// animation-part replacements — that lives inside both CreateObject
+    /// (initial spawn) and ObjDescEvent (0xF625 appearance update).
+    ///
+    /// <para>Layout: byte marker (0x11), byte subPaletteCount, byte
+    /// textureChangeCount, byte animPartChangeCount. Then:</para>
+    /// <list type="bullet">
+    ///   <item>BasePaletteId (PackedDword of palette type), only present when subPaletteCount &gt; 0</item>
+    ///   <item>SubPalettes[subPaletteCount]: PackedDword id + byte offset + byte length</item>
+    ///   <item>TextureChanges[textureChangeCount]: byte partIndex + PackedDword oldTex + PackedDword newTex</item>
+    ///   <item>AnimPartChanges[animPartChangeCount]: byte partIndex + PackedDword newModelId</item>
+    ///   <item>4-byte alignment pad</item>
+    /// </list>
+    ///
+    /// <para>Throws <see cref="FormatException"/> on truncated input —
+    /// callers wrap in try/catch and convert to a null result. Advances
+    /// <paramref name="pos"/> past the alignment pad so the caller can
+    /// continue reading the next field.</para>
+    /// </summary>
+    public static ModelData ReadModelData(ReadOnlySpan<byte> body, ref int pos)
+    {
+        if (body.Length - pos < 4) throw new FormatException("truncated ModelData header");
+        byte _marker = body[pos]; pos += 1;
+        byte subPaletteCount = body[pos]; pos += 1;
+        byte textureChangeCount = body[pos]; pos += 1;
+        byte animPartChangeCount = body[pos]; pos += 1;
+
+        uint? basePaletteId = null;
+        if (subPaletteCount > 0)
+            basePaletteId = ReadPackedDwordOfKnownType(body, ref pos, PaletteTypePrefix);
+
+        var subPalettes = subPaletteCount == 0
+            ? (IReadOnlyList<SubPaletteSwap>)Array.Empty<SubPaletteSwap>()
+            : new SubPaletteSwap[subPaletteCount];
+        for (int i = 0; i < subPaletteCount; i++)
+        {
+            uint subPalId = ReadPackedDwordOfKnownType(body, ref pos, PaletteTypePrefix);
+            if (body.Length - pos < 2) throw new FormatException("truncated SubPaletteSwap");
+            byte offset = body[pos]; pos += 1;
+            byte length = body[pos]; pos += 1;
+            ((SubPaletteSwap[])subPalettes)[i] = new SubPaletteSwap(subPalId, offset, length);
+        }
+
+        var textureChanges = textureChangeCount == 0
+            ? (IReadOnlyList<TextureChange>)Array.Empty<TextureChange>()
+            : new TextureChange[textureChangeCount];
+        for (int i = 0; i < textureChangeCount; i++)
+        {
+            if (body.Length - pos < 1) throw new FormatException("truncated TextureChange");
+            byte partIndex = body[pos]; pos += 1;
+            uint oldTex = ReadPackedDwordOfKnownType(body, ref pos, SurfaceTextureTypePrefix);
+            uint newTex = ReadPackedDwordOfKnownType(body, ref pos, SurfaceTextureTypePrefix);
+            ((TextureChange[])textureChanges)[i] = new TextureChange(partIndex, oldTex, newTex);
+        }
+
+        // ACE writes NewModelId via WritePackedDwordOfKnownType(0x01000000)
+        // which strips the high-byte type if present before packing.
+        // ReadPackedDwordOfKnownType ORs it back on read.
+        var animParts = animPartChangeCount == 0
+            ? (IReadOnlyList<AnimPartChange>)Array.Empty<AnimPartChange>()
+            : new AnimPartChange[animPartChangeCount];
+        for (int i = 0; i < animPartChangeCount; i++)
+        {
+            if (body.Length - pos < 1) throw new FormatException("truncated AnimPartChange");
+            byte partIndex = body[pos]; pos += 1;
+            uint newModelId = ReadPackedDwordOfKnownType(body, ref pos, GfxObjTypePrefix);
+            ((AnimPartChange[])animParts)[i] = new AnimPartChange(partIndex, newModelId);
+        }
+
+        AlignTo4(ref pos);
+        return new ModelData(basePaletteId, subPalettes, textureChanges, animParts);
+    }
+
     private static uint ReadU32(ReadOnlySpan<byte> source, ref int pos)
     {
         if (source.Length - pos < 4) throw new FormatException("truncated u32");

src/AcDream.Core.Net/Messages/ObjDescEvent.cs

@@ -0,0 +1,74 @@
+using System.Buffers.Binary;
+
+namespace AcDream.Core.Net.Messages;
+
+/// <summary>
+/// Inbound <c>ObjDescEvent</c> GameMessage (opcode <c>0xF625</c>). ACE
+/// broadcasts this whenever a creature/player's appearance changes after
+/// the initial <see cref="CreateObject"/> spawn — equip / unequip
+/// (Creature_Equipment.cs:365), tailoring (Tailoring.cs:504), recipe
+/// results (RecipeManager.cs:403), character-option toggles. Skunkwors
+/// protocol docs: "F625: Change Model — Sent whenever a character changes
+/// their clothes. It contains the entire description of what they're
+/// wearing (and possibly their facial features as well). This message is
+/// only sent for changes; when the character is first created, the body
+/// of this message is included inside the creation message."
+///
+/// <para>Retail handles it via <c>SmartBox::HandleObjDescEvent</c>
+/// (named-retail symbol 0x453340). acdream silently dropped it through
+/// 2026-05-06 — the bug was that retail-driven characters observed from
+/// acdream rendered with the wrong skin/hair palettes because the
+/// follow-up appearance updates were never applied.</para>
+///
+/// <para>Wire layout (ACE WorldObject_Networking.cs:48-54
+/// <c>SerializeUpdateModelData</c>):</para>
+/// <list type="bullet">
+///   <item>u32 opcode (0xF625)</item>
+///   <item>u32 guid — target object</item>
+///   <item>ModelData block — see <see cref="CreateObject.ReadModelData"/></item>
+///   <item>u32 instanceSequence</item>
+///   <item>u32 visualDescSequence</item>
+/// </list>
+/// </summary>
+public static class ObjDescEvent
+{
+    public const uint Opcode = 0xF625u;
+
+    /// <summary>
+    /// One ObjDescEvent: target guid + the new ModelData. Sequence
+    /// counters are read but not surfaced (subscribers don't need them
+    /// — the event always carries the full new appearance).
+    /// </summary>
+    public readonly record struct Parsed(uint Guid, CreateObject.ModelData ModelData);
+
+    /// <summary>
+    /// Parse an ObjDescEvent body (must start with the 4-byte opcode).
+    /// Returns null on truncation or wrong opcode.
+    /// </summary>
+    public static Parsed? TryParse(ReadOnlySpan<byte> body)
+    {
+        try
+        {
+            int pos = 0;
+            if (body.Length - pos < 4) return null;
+            uint opcode = BinaryPrimitives.ReadUInt32LittleEndian(body.Slice(pos));
+            pos += 4;
+            if (opcode != Opcode) return null;
+
+            if (body.Length - pos < 4) return null;
+            uint guid = BinaryPrimitives.ReadUInt32LittleEndian(body.Slice(pos));
+            pos += 4;
+
+            var modelData = CreateObject.ReadModelData(body, ref pos);
+
+            // Trailing instanceSeq + visualDescSeq are read for completeness
+            // but not surfaced — subscribers re-render unconditionally on
+            // every event since each carries the full appearance.
+            return new Parsed(guid, modelData);
+        }
+        catch
+        {
+            return null;
+        }
+    }
+}

src/AcDream.Core.Net/WorldSession.cs

@@ -139,6 +139,20 @@ public sealed class WorldSession : IDisposable
     /// </summary>
     public event Action<uint>? TeleportStarted;
 
+    /// <summary>
+    /// Fires when the server broadcasts an <c>ObjDescEvent (0xF625)</c> —
+    /// a creature/player's appearance changed after the initial CreateObject
+    /// (equip / unequip / tailoring / recipe result / character option toggle).
+    /// Subscribers re-apply the new <c>ModelData</c> to the existing entity:
+    /// AnimPartChanges replace mesh refs, TextureChanges update per-part
+    /// surface texture overrides, and SubPalettes rebuild the palette
+    /// override (the channel that carries skin/hair tone). Without this,
+    /// retail-driven characters observed from acdream end up "stuck" at
+    /// whatever appearance was in their first CreateObject — see issue
+    /// notes in commit history around 2026-05-06.
+    /// </summary>
+    public event Action<ObjDescEvent.Parsed>? AppearanceUpdated;
+
     /// <summary>
     /// Phase H.1: fires when a local or ranged speech message (0x02BB /
     /// 0x02BC) is received. Subscribers typically feed these into a
@@ -314,12 +328,18 @@ public sealed class WorldSession : IDisposable
     private readonly FragmentAssembler _assembler = new();
 
     // Issue #5 diagnostics (env-var-gated):
-    //   ACDREAM_DUMP_OPCODES=1 → log first occurrence of each unhandled opcode
-    //   ACDREAM_DUMP_VITALS=1  → log every PrivateUpdateVital(Current) parse
+    //   ACDREAM_DUMP_OPCODES=1     → log first occurrence of each unhandled opcode
+    //   ACDREAM_DUMP_VITALS=1      → log every PrivateUpdateVital(Current) parse
+    //   ACDREAM_DUMP_APPEARANCE=1  → log every 0xF625 ObjDescEvent + 0xF7DB UpdateObject
+    //                                with body len, target guid, hex preview. Used to
+    //                                debug remote-player appearance asymmetry (retail
+    //                                observer in acdream renders wrong skin/hair).
     private static readonly bool DumpOpcodesEnabled =
         Environment.GetEnvironmentVariable("ACDREAM_DUMP_OPCODES") == "1";
     private static readonly bool DumpVitalsEnabled =
         Environment.GetEnvironmentVariable("ACDREAM_DUMP_VITALS") == "1";
+    private static readonly bool DumpAppearanceEnabled =
+        Environment.GetEnvironmentVariable("ACDREAM_DUMP_APPEARANCE") == "1";
     private readonly System.Collections.Generic.HashSet<uint> _seenUnhandledOpcodes = new();
 
     private IsaacRandom? _inboundIsaac;
@@ -861,6 +881,36 @@ public sealed class WorldSession : IDisposable
                 _teleportSequence = sequence;  // track for outbound movement messages
                 TeleportStarted?.Invoke(sequence);
             }
+            else if (op == ObjDescEvent.Opcode)
+            {
+                // 0xF625 ObjDescEvent — per-entity appearance update. ACE
+                // broadcasts on equip/unequip/tailoring/recipe/option-change
+                // (Creature_Equipment.cs:365, Tailoring.cs:504,
+                // RecipeManager.cs:403, GameActionSetSingleCharacterOption.cs:27).
+                // Retail handler: SmartBox::HandleObjDescEvent (named-retail
+                // 0x453340). Body layout: u32 opcode | u32 guid | ModelData |
+                // u32 instanceSeq | u32 visualDescSeq.
+                var parsed = ObjDescEvent.TryParse(body);
+                if (parsed is not null)
+                {
+                    if (DumpAppearanceEnabled)
+                    {
+                        var md = parsed.Value.ModelData;
+                        Console.WriteLine($"appearance: 0xF625 guid=0x{parsed.Value.Guid:X8} basePal=0x{(md.BasePaletteId ?? 0):X8} subPals={md.SubPalettes.Count} texChanges={md.TextureChanges.Count} animParts={md.AnimPartChanges.Count}");
+                        foreach (var sp in md.SubPalettes)
+                            Console.WriteLine($"  SP id=0x{sp.SubPaletteId:X8} offset={sp.Offset} length={sp.Length}");
+                        foreach (var tc in md.TextureChanges)
+                            Console.WriteLine($"  TC part={tc.PartIndex:D2} oldTex=0x{tc.OldTexture:X8} -> newTex=0x{tc.NewTexture:X8}");
+                        foreach (var apc in md.AnimPartChanges)
+                            Console.WriteLine($"  APC part={apc.PartIndex:D2} -> gfx=0x{apc.NewModelId:X8}");
+                    }
+                    AppearanceUpdated?.Invoke(parsed.Value);
+                }
+                else if (DumpAppearanceEnabled)
+                {
+                    Console.WriteLine($"appearance: 0xF625 PARSE FAILED body.len={body.Length}");
+                }
+            }
             else if (DumpOpcodesEnabled)
             {
                 // ACDREAM_DUMP_OPCODES=1 — emit a one-line trace per

src/AcDream.Core/Meshing/GfxObjDegradeResolver.cs

@@ -0,0 +1,144 @@
+using DatReaderWriter;
+using DatReaderWriter.DBObjs;
+using DatReaderWriter.Enums;
+
+namespace AcDream.Core.Meshing;
+
+/// <summary>
+/// Resolve a base GfxObj id to its retail "close-detail" mesh by walking
+/// the <c>DIDDegrade</c> table to <c>Degrades[0]</c>.
+///
+/// <para>
+/// <b>Why this exists (Issue #47).</b> Many AC GfxObjs — most notably
+/// humanoid body parts — store the LOW-detail mesh as the GfxObj that
+/// the Setup or AnimPartChange references. The high-detail mesh used
+/// for close/player rendering is reached indirectly: the base GfxObj's
+/// <c>HasDIDDegrade</c> flag is set, <c>DIDDegrade</c> points at a
+/// <see cref="GfxObjDegradeInfo"/>, and <see cref="GfxObjInfo.Id"/> at
+/// <c>Degrades[0]</c> is the close-detail variant. Drawing the base
+/// GfxObj id directly produces the LOD-3 mesh — visibly bulky and
+/// detail-less — which is exactly what acdream and ACViewer were both
+/// rendering for humanoid body parts before this fix.
+/// </para>
+///
+/// <para>
+/// <b>Concrete example.</b> The Aluvian Male upper-arm GfxObj
+/// <c>0x01000055</c> is a 14-vertex / 17-poly low-detail stub. Its
+/// degrade table <c>0x110006D0</c> points at <c>0x01001795</c>, the
+/// 32-vertex / 60-poly close-detail mesh that carries the bicep /
+/// deltoid / shoulder geometry retail draws on the player. Same story
+/// for the lower arm <c>0x01000056 → 0x0100178F</c> and matching
+/// heritage variants (<c>0x010004BF → 0x010017A8</c>,
+/// <c>0x010004BD → 0x010017A7</c>, <c>0x010004B7 → 0x0100179A</c>).
+/// </para>
+///
+/// <para>
+/// <b>Retail flow (named-retail decomp).</b>
+/// <list type="bullet">
+///   <item>
+///     <c>acclient!CPhysicsPart::LoadGfxObjArray</c> at <c>0x0050DCF0</c>
+///     loads the base GfxObj solely to discover <c>DIDDegrade</c>; if
+///     a <see cref="GfxObjDegradeInfo"/> exists, retail loads each entry
+///     in <c>Degrades</c> into the part's render array.
+///   </item>
+///   <item>
+///     <c>acclient!CPhysicsPart::UpdateViewerDistance</c> at
+///     <c>0x0050E030</c> picks <c>deg_level</c> per part by distance.
+///     For close / player rendering <c>deg_level == 0</c>.
+///   </item>
+///   <item>
+///     <c>acclient!CPhysicsPart::Draw</c> at <c>0x0050D7A0</c>
+///     draws <c>gfxobj[deg_level]</c>.
+///   </item>
+/// </list>
+/// </para>
+///
+/// <para>
+/// We don't yet have distance-based LOD plumbing, so this resolver
+/// always returns slot 0 (the close-detail mesh). That's correct for
+/// player + nearby NPC rendering; far-distance LOD is a future concern.
+/// </para>
+/// </summary>
+public static class GfxObjDegradeResolver
+{
+    /// <summary>
+    /// DatCollection-backed convenience overload. Production callers use
+    /// this; tests use the callback overload below for easy fakes.
+    /// </summary>
+    public static bool TryResolveCloseGfxObj(
+        DatCollection dats,
+        uint gfxObjId,
+        out uint resolvedId,
+        out GfxObj? resolvedGfxObj)
+        => TryResolveCloseGfxObj(
+            id => dats.Get<GfxObj>(id),
+            id => dats.Get<GfxObjDegradeInfo>(id),
+            gfxObjId,
+            out resolvedId,
+            out resolvedGfxObj);
+
+    /// <summary>
+    /// Loader-callback overload. Returns the close-detail GfxObj id and
+    /// loaded object when a degrade table is present, otherwise the
+    /// base id and base GfxObj.
+    /// </summary>
+    /// <param name="getGfxObj">
+    /// Lookup for a GfxObj by id. May return null when not found.
+    /// </param>
+    /// <param name="getDegradeInfo">
+    /// Lookup for a GfxObjDegradeInfo by id. May return null.
+    /// </param>
+    /// <param name="gfxObjId">Base GfxObj id (post-AnimPartChange).</param>
+    /// <param name="resolvedId">
+    /// The id to actually render. Same as <paramref name="gfxObjId"/>
+    /// when no degrade table exists; <c>Degrades[0].Id</c> when it does.
+    /// </param>
+    /// <param name="resolvedGfxObj">
+    /// The loaded GfxObj for <paramref name="resolvedId"/>, cached so
+    /// callers don't have to re-read.
+    /// </param>
+    /// <returns>
+    /// <c>true</c> if a usable GfxObj was resolved (either base or
+    /// degrade slot 0 loaded). <c>false</c> only when the base GfxObj
+    /// itself was missing — caller should drop this part.
+    /// </returns>
+    public static bool TryResolveCloseGfxObj(
+        Func<uint, GfxObj?> getGfxObj,
+        Func<uint, GfxObjDegradeInfo?> getDegradeInfo,
+        uint gfxObjId,
+        out uint resolvedId,
+        out GfxObj? resolvedGfxObj)
+    {
+        var gfxObj = getGfxObj(gfxObjId);
+        if (gfxObj is null)
+        {
+            resolvedId = gfxObjId;
+            resolvedGfxObj = null;
+            return false;
+        }
+
+        // Default: base mesh stays selected unless the degrade table
+        // resolves cleanly. Every fallback below leaves these set.
+        resolvedId = gfxObjId;
+        resolvedGfxObj = gfxObj;
+
+        if (!gfxObj.Flags.HasFlag(GfxObjFlags.HasDIDDegrade) || gfxObj.DIDDegrade == 0)
+            return true;
+
+        var degradeInfo = getDegradeInfo(gfxObj.DIDDegrade);
+        if (degradeInfo is null || degradeInfo.Degrades.Count == 0)
+            return true;
+
+        uint closeId = (uint)degradeInfo.Degrades[0].Id;
+        if (closeId == 0)
+            return true;
+
+        var closeGfxObj = getGfxObj(closeId);
+        if (closeGfxObj is null)
+            return true;
+
+        resolvedId = closeId;
+        resolvedGfxObj = closeGfxObj;
+        return true;
+    }
+}

src/AcDream.Core/Physics/AnimationSequencer.cs

@@ -254,6 +254,39 @@ public sealed class AnimationSequencer
     public int QueueCount => _queue.Count;
     public bool HasCurrentNode => _currNode != null;
 
+    /// <summary>
+    /// Diagnostic snapshot of <c>_currNode</c>'s identity + frame state, for
+    /// the per-tick CURRNODE log line in <c>GameWindow.TickAnimations</c>.
+    /// Lets the caller see whether the actual node being read by Advance /
+    /// BuildBlendedFrame is what SetCycle was supposed to leave it on.
+    /// AnimRefHash uses object-identity hashing on the Animation reference
+    /// so different Walk vs Run anim resources can be distinguished even
+    /// without exposing the full Animation type.
+    /// </summary>
+    public (int AnimRefHash, bool IsLooping, double Framerate, int StartFrame, int EndFrame, double FramePosition, int QueueCount) CurrentNodeDiag
+    {
+        get
+        {
+            if (_currNode is null)
+                return (0, false, 0.0, 0, 0, 0.0, _queue.Count);
+            var n = _currNode.Value;
+            int hash = System.Runtime.CompilerServices.RuntimeHelpers.GetHashCode(n.Anim);
+            return (hash, n.IsLooping, n.Framerate, n.StartFrame, n.EndFrame, _framePosition, _queue.Count);
+        }
+    }
+
+    /// <summary>
+    /// Diagnostic: the AnimRefHash for the FIRST cyclic node in the queue
+    /// (i.e., what SetCycle is trying to land us on for a locomotion cycle).
+    /// Compare against <see cref="CurrentNodeDiag"/>'s AnimRefHash to see
+    /// whether <c>_currNode</c> is actually pointing at the new cycle or
+    /// something stale.
+    /// </summary>
+    public int FirstCyclicAnimRefHash =>
+        _firstCyclic is null
+            ? 0
+            : System.Runtime.CompilerServices.RuntimeHelpers.GetHashCode(_firstCyclic.Value.Anim);
+
     // ── Private state ────────────────────────────────────────────────────────
 
     private readonly Setup _setup;
@@ -525,7 +558,59 @@ public sealed class AnimationSequencer
         // newly-added node; if preEnqueueTail was null (queue was empty
         // before enqueue), the first new node is _queue.First.
         var firstNew = preEnqueueTail is null ? _queue.First : preEnqueueTail.Next;
-        if (firstNew is not null)
+
+        // #39 Fix B (2026-05-06): for direct cyclic-locomotion →
+        // cyclic-locomotion transitions (Walk↔Run on Shift toggle,
+        // W↔S direct flip, A↔D, Forward↔Strafe), land _currNode on
+        // the new CYCLE (_firstCyclic), NOT on the link (firstNew),
+        // and remove the just-enqueued link from the queue.
+        //
+        // Why: the transition link's drain time (~100–300 ms at
+        // Framerate 30 × link runSpeed) gets restarted before it can
+        // end if the user toggles Shift faster than that. _currNode
+        // sits on a fresh link every UM and Advance never reaches
+        // the cycle. User observes "blips forward in walking
+        // animation" — what they're seeing is the link's
+        // interpolation pose, never the new cycle.
+        //
+        // Conditional on BOTH old AND new being locomotion cycles to
+        // avoid regressing the cases where the link IS the right
+        // animation:
+        //   - Idle (Ready) → any cycle: link is the wind-up pose
+        //   - Falling → Ready: landing animation
+        //   - Ready → Sitting/Crouching: pose-change links
+        //   - Combat substates (attack/parry/ready transitions)
+        // Commit c06b6c5 (reverted in a2ae2ae) demonstrated that
+        // unconditionally skipping the link breaks all of these.
+        //
+        // Retail reference: cdb live trace 2026-05-03 of a Walk→Run
+        // direct transition logged
+        //   add_to_queue(45000005, looping=1)   walk
+        //   add_to_queue(44000007, looping=1)   run
+        // with truncate_animation_list never firing — i.e. retail
+        // appends the new cycle directly without a separate link
+        // enqueue or visible link pose for cyclic→cyclic. Our
+        // structural mismatch was always enqueueing link+cycle and
+        // forcing _currNode onto the link; this fix matches retail's
+        // observed semantics for the locomotion subset.
+        bool prevIsLocomotion = IsLocomotionCycleLowByte(CurrentMotion & 0xFFu);
+        bool newIsLocomotion = IsLocomotionCycleLowByte(motion & 0xFFu);
+        if (prevIsLocomotion && newIsLocomotion && _firstCyclic is not null)
+        {
+            // Drop the just-enqueued link node (firstNew) from the
+            // queue if it's distinct from the cycle — nothing should
+            // ever play it, and leaving stale non-cyclic nodes ahead
+            // of _currNode contributes to the unbounded queue growth
+            // observed in [SCFULL] (qCount climbing past 49 over
+            // ~30 transitions).
+            if (firstNew is not null && firstNew != _firstCyclic)
+            {
+                _queue.Remove(firstNew);
+            }
+            _currNode = _firstCyclic;
+            _framePosition = _firstCyclic.Value.GetStartFramePosition();
+        }
+        else if (firstNew is not null)
         {
             _currNode = firstNew;
             _framePosition = _currNode.Value.GetStartFramePosition();
@@ -1384,6 +1469,20 @@ public sealed class AnimationSequencer
         return result;
     }
 
+    /// <summary>
+    /// True if the given motion-low-byte names a locomotion cycle —
+    /// WalkForward (0x05), WalkBackward (0x06), RunForward (0x07),
+    /// SideStepRight (0x0F), or SideStepLeft (0x10).
+    /// Used by <see cref="SetCycle"/> to recognise cyclic→cyclic
+    /// direct transitions and bypass the transition link in that case
+    /// (retail's observed add_to_queue semantics).
+    /// </summary>
+    private static bool IsLocomotionCycleLowByte(uint lowByte)
+    {
+        return lowByte == 0x05u || lowByte == 0x06u || lowByte == 0x07u
+            || lowByte == 0x0Fu || lowByte == 0x10u;
+    }
+
     /// <summary>
     /// Quaternion slerp matching the retail client's <c>FUN_005360d0</c>
     /// (chunk_00530000.c:4799-4846):

tests/AcDream.Core.Net.Tests/Messages/ObjDescEventTests.cs

@@ -0,0 +1,153 @@
+using System.Buffers.Binary;
+using AcDream.Core.Net.Messages;
+
+namespace AcDream.Core.Net.Tests.Messages;
+
+public sealed class ObjDescEventTests
+{
+    [Fact]
+    public void TryParse_RejectsWrongOpcode()
+    {
+        byte[] body = new byte[16];
+        BinaryPrimitives.WriteUInt32LittleEndian(body, 0xF745u); // CreateObject opcode
+        Assert.Null(ObjDescEvent.TryParse(body));
+    }
+
+    [Fact]
+    public void TryParse_RejectsTruncatedBody()
+    {
+        Assert.Null(ObjDescEvent.TryParse(new byte[3]));
+    }
+
+    /// <summary>
+    /// Round-trip a synthesized body: opcode + guid + ModelData (3 SubPalettes,
+    /// 4 TextureChanges, 0 AnimPartChanges — same shape as the captured retail
+    /// 152-byte +Je ObjDescEvent body) + trailing sequence pair. Verifies the
+    /// parser surfaces the same fields the writer wrote.
+    /// </summary>
+    [Fact]
+    public void TryParse_SynthesizedBody_ExtractsGuidAndModelData()
+    {
+        // Build a body matching the wire shape we see from ACE.
+        var bytes = new List<byte>();
+        AppendU32(bytes, ObjDescEvent.Opcode);
+        AppendU32(bytes, 0x50000001u); // target guid
+
+        // ModelData header: marker, subPalCount, texCount, animPartCount.
+        bytes.Add(0x11);
+        bytes.Add(3); // subPalCount
+        bytes.Add(4); // texChangeCount
+        bytes.Add(0); // animPartCount
+
+        // BasePaletteId (palette type prefix stripped before packing).
+        AppendPackedDword(bytes, 0x0400007Eu, 0x04000000u);
+
+        // SubPalettes — three skin/hair-style overlays at varied offsets.
+        AppendPackedDword(bytes, 0x04001FE3u, 0x04000000u);
+        bytes.Add(24); bytes.Add(8);
+        AppendPackedDword(bytes, 0x040002BAu, 0x04000000u);
+        bytes.Add(0); bytes.Add(24);
+        AppendPackedDword(bytes, 0x040002BCu, 0x04000000u);
+        bytes.Add(32); bytes.Add(8);
+
+        // TextureChanges — four part textures.
+        for (byte partIdx = 0; partIdx < 4; partIdx++)
+        {
+            bytes.Add(partIdx);
+            AppendPackedDword(bytes, 0x05000100u + partIdx, 0x05000000u);
+            AppendPackedDword(bytes, 0x05000200u + partIdx, 0x05000000u);
+        }
+
+        // 4-byte align after AnimPartChanges (none here, so just align).
+        while (bytes.Count % 4 != 0) bytes.Add(0);
+
+        // Trailing instance + visual-desc sequences (consumed but ignored).
+        AppendU32(bytes, 0x12345678u);
+        AppendU32(bytes, 0x9ABCDEF0u);
+
+        var parsed = ObjDescEvent.TryParse(bytes.ToArray());
+
+        Assert.NotNull(parsed);
+        Assert.Equal(0x50000001u, parsed!.Value.Guid);
+
+        var md = parsed.Value.ModelData;
+        Assert.Equal(0x0400007Eu, md.BasePaletteId);
+        Assert.Equal(3, md.SubPalettes.Count);
+        Assert.Equal(0x04001FE3u, md.SubPalettes[0].SubPaletteId);
+        Assert.Equal(24, md.SubPalettes[0].Offset);
+        Assert.Equal(8, md.SubPalettes[0].Length);
+        Assert.Equal(0x040002BAu, md.SubPalettes[1].SubPaletteId);
+        Assert.Equal(0, md.SubPalettes[1].Offset);
+        Assert.Equal(24, md.SubPalettes[1].Length);
+
+        Assert.Equal(4, md.TextureChanges.Count);
+        Assert.Equal(0, md.TextureChanges[0].PartIndex);
+        Assert.Equal(0x05000100u, md.TextureChanges[0].OldTexture);
+        Assert.Equal(0x05000200u, md.TextureChanges[0].NewTexture);
+        Assert.Equal(3, md.TextureChanges[3].PartIndex);
+
+        Assert.Empty(md.AnimPartChanges);
+    }
+
+    /// <summary>
+    /// Confirms ReadModelData (the shared helper) round-trips identically
+    /// when called from CreateObject and from ObjDescEvent — same bytes,
+    /// same parsed output. Guards against the two callers drifting.
+    /// </summary>
+    [Fact]
+    public void ReadModelData_SameOutputFromBothCallers()
+    {
+        // Bare ModelData block — used as a substring in both messages.
+        var modelDataBytes = new List<byte>();
+        modelDataBytes.Add(0x11);
+        modelDataBytes.Add(1); // subPalCount
+        modelDataBytes.Add(0); // texCount
+        modelDataBytes.Add(0); // animPartCount
+        AppendPackedDword(modelDataBytes, 0x0400007Eu, 0x04000000u);
+        AppendPackedDword(modelDataBytes, 0x04001084u, 0x04000000u);
+        modelDataBytes.Add(80); modelDataBytes.Add(12);
+        while (modelDataBytes.Count % 4 != 0) modelDataBytes.Add(0);
+
+        ReadOnlySpan<byte> span = modelDataBytes.ToArray();
+        int pos = 0;
+        var md = CreateObject.ReadModelData(span, ref pos);
+
+        Assert.Equal(0x0400007Eu, md.BasePaletteId);
+        Assert.Single(md.SubPalettes);
+        Assert.Equal(0x04001084u, md.SubPalettes[0].SubPaletteId);
+        Assert.Equal(80, md.SubPalettes[0].Offset);
+        Assert.Equal(12, md.SubPalettes[0].Length);
+    }
+
+    private static void AppendU32(List<byte> dest, uint value)
+    {
+        Span<byte> tmp = stackalloc byte[4];
+        BinaryPrimitives.WriteUInt32LittleEndian(tmp, value);
+        dest.AddRange(tmp.ToArray());
+    }
+
+    /// <summary>
+    /// Mirror of ACE's WritePackedDwordOfKnownType: strip the type prefix
+    /// if it matches <paramref name="knownType"/>, then write as a 16- or
+    /// 32-bit packed value.
+    /// </summary>
+    private static void AppendPackedDword(List<byte> dest, uint value, uint knownType)
+    {
+        uint packed = (value & 0xFF000000u) == knownType ? (value & ~knownType) : value;
+        if (packed <= 0x7FFFu)
+        {
+            Span<byte> tmp = stackalloc byte[2];
+            BinaryPrimitives.WriteUInt16LittleEndian(tmp, (ushort)packed);
+            dest.AddRange(tmp.ToArray());
+        }
+        else
+        {
+            ushort high = (ushort)((packed >> 16) | 0x8000);
+            ushort low  = (ushort)(packed & 0xFFFFu);
+            Span<byte> tmp = stackalloc byte[4];
+            BinaryPrimitives.WriteUInt16LittleEndian(tmp, high);
+            BinaryPrimitives.WriteUInt16LittleEndian(tmp.Slice(2), low);
+            dest.AddRange(tmp.ToArray());
+        }
+    }
+}

tests/AcDream.Core.Tests/Meshing/GfxObjDegradeResolverTests.cs

@@ -0,0 +1,182 @@
+using AcDream.Core.Meshing;
+using DatReaderWriter.DBObjs;
+using DatReaderWriter.Enums;
+using DatReaderWriter.Types;
+
+namespace AcDream.Core.Tests.Meshing;
+
+/// <summary>
+/// Unit tests for <see cref="GfxObjDegradeResolver"/>. The resolver is
+/// the Issue #47 fix: route a base GfxObj id to its retail close-detail
+/// mesh via the DIDDegrade table's slot 0. Tests use the callback
+/// overload so we can stand up tiny in-memory fixtures without dragging
+/// in a real DatCollection.
+/// </summary>
+public class GfxObjDegradeResolverTests
+{
+    /// <summary>
+    /// When the base GfxObj has no degrade table (HasDIDDegrade flag
+    /// clear), the resolver returns the base id unchanged.
+    /// </summary>
+    [Fact]
+    public void NoDegradeTable_ReturnsBaseMesh()
+    {
+        const uint baseId = 0x01001212u;
+        var baseGfx = new GfxObj { Flags = 0, DIDDegrade = 0 };
+        var gfxObjs = new Dictionary<uint, GfxObj> { [baseId] = baseGfx };
+
+        bool ok = GfxObjDegradeResolver.TryResolveCloseGfxObj(
+            id => gfxObjs.GetValueOrDefault(id),
+            _ => null,
+            baseId,
+            out uint resolvedId,
+            out var resolvedGfx);
+
+        Assert.True(ok);
+        Assert.Equal(baseId, resolvedId);
+        Assert.Same(baseGfx, resolvedGfx);
+    }
+
+    /// <summary>
+    /// When the base GfxObj has a populated DIDDegrade table, the
+    /// resolver returns Degrades[0].Id and its loaded GfxObj — the
+    /// close-detail mesh retail draws for nearby objects.
+    /// </summary>
+    [Fact]
+    public void ValidDegradeTable_ReturnsSlotZero()
+    {
+        const uint baseId = 0x01000055u;       // low-detail Aluvian Male upper arm
+        const uint degradeInfoId = 0x110006D0u;
+        const uint closeId = 0x01001795u;      // retail close-detail variant
+
+        var baseGfx = new GfxObj
+        {
+            Flags = GfxObjFlags.HasDIDDegrade,
+            DIDDegrade = degradeInfoId,
+        };
+        var closeGfx = new GfxObj { Flags = 0 };
+        var degradeInfo = new GfxObjDegradeInfo
+        {
+            Degrades = { new GfxObjInfo { Id = closeId } },
+        };
+
+        var gfxObjs = new Dictionary<uint, GfxObj>
+        {
+            [baseId] = baseGfx,
+            [closeId] = closeGfx,
+        };
+        var degradeInfos = new Dictionary<uint, GfxObjDegradeInfo>
+        {
+            [degradeInfoId] = degradeInfo,
+        };
+
+        bool ok = GfxObjDegradeResolver.TryResolveCloseGfxObj(
+            id => gfxObjs.GetValueOrDefault(id),
+            id => degradeInfos.GetValueOrDefault(id),
+            baseId,
+            out uint resolvedId,
+            out var resolvedGfx);
+
+        Assert.True(ok);
+        Assert.Equal(closeId, resolvedId);
+        Assert.Same(closeGfx, resolvedGfx);
+    }
+
+    /// <summary>
+    /// If the degrade table references a GfxObj that isn't present in
+    /// the dat (corrupt / partial dat), the resolver falls back to the
+    /// base mesh rather than returning null. Better to render the
+    /// low-detail variant than nothing at all.
+    /// </summary>
+    [Fact]
+    public void MissingSlotZeroMesh_FallsBackToBase()
+    {
+        const uint baseId = 0x01000055u;
+        const uint degradeInfoId = 0x110006D0u;
+        const uint missingCloseId = 0xDEADBEEFu;
+
+        var baseGfx = new GfxObj
+        {
+            Flags = GfxObjFlags.HasDIDDegrade,
+            DIDDegrade = degradeInfoId,
+        };
+        var degradeInfo = new GfxObjDegradeInfo
+        {
+            Degrades = { new GfxObjInfo { Id = missingCloseId } },
+        };
+        var gfxObjs = new Dictionary<uint, GfxObj> { [baseId] = baseGfx };
+        var degradeInfos = new Dictionary<uint, GfxObjDegradeInfo>
+        {
+            [degradeInfoId] = degradeInfo,
+        };
+
+        bool ok = GfxObjDegradeResolver.TryResolveCloseGfxObj(
+            id => gfxObjs.GetValueOrDefault(id),
+            id => degradeInfos.GetValueOrDefault(id),
+            baseId,
+            out uint resolvedId,
+            out var resolvedGfx);
+
+        Assert.True(ok);
+        Assert.Equal(baseId, resolvedId);
+        Assert.Same(baseGfx, resolvedGfx);
+    }
+
+    /// <summary>
+    /// Empty Degrades list (table present but no entries) falls back
+    /// to base. Mirrors retail's "no LOD entries → just draw the base"
+    /// behavior.
+    /// </summary>
+    [Fact]
+    public void EmptyDegradesList_FallsBackToBase()
+    {
+        const uint baseId = 0x01000055u;
+        const uint degradeInfoId = 0x110006D0u;
+
+        var baseGfx = new GfxObj
+        {
+            Flags = GfxObjFlags.HasDIDDegrade,
+            DIDDegrade = degradeInfoId,
+        };
+        var degradeInfo = new GfxObjDegradeInfo();   // empty Degrades
+
+        var gfxObjs = new Dictionary<uint, GfxObj> { [baseId] = baseGfx };
+        var degradeInfos = new Dictionary<uint, GfxObjDegradeInfo>
+        {
+            [degradeInfoId] = degradeInfo,
+        };
+
+        bool ok = GfxObjDegradeResolver.TryResolveCloseGfxObj(
+            id => gfxObjs.GetValueOrDefault(id),
+            id => degradeInfos.GetValueOrDefault(id),
+            baseId,
+            out uint resolvedId,
+            out var resolvedGfx);
+
+        Assert.True(ok);
+        Assert.Equal(baseId, resolvedId);
+        Assert.Same(baseGfx, resolvedGfx);
+    }
+
+    /// <summary>
+    /// When the base GfxObj itself is missing from the dat, the
+    /// resolver returns false so the caller can drop the part rather
+    /// than trying to render a null mesh.
+    /// </summary>
+    [Fact]
+    public void MissingBaseGfxObj_ReturnsFalse()
+    {
+        const uint baseId = 0xDEADBEEFu;
+
+        bool ok = GfxObjDegradeResolver.TryResolveCloseGfxObj(
+            _ => null,
+            _ => null,
+            baseId,
+            out uint resolvedId,
+            out var resolvedGfx);
+
+        Assert.False(ok);
+        Assert.Equal(baseId, resolvedId);
+        Assert.Null(resolvedGfx);
+    }
+}