docs(B.4b): final-review polish — file #59 #60 follow-ups + handoff correction

Final whole-branch code review (Opus) surfaced two Important post-merge
follow-ups + a one-word inaccuracy in the handoff doc:

- #59: tighten WorldPicker per-entity Setup.Radius (M1-deferred; the
  ServerGuid==0 invariant is load-bearing and worth documenting before
  L.2d's CBuildingObj port lands).
- #60: port retail's obstruction_ethereal downstream path so combat-HUD
  contact reporting works for ethereal creatures (M2-combat).
- handoff: corrected "Added a _entitiesByServerGuid reverse-lookup" to
  "Used the pre-existing _entitiesByServerGuid" — the dict has existed
  since Phase 6.6/6.7; slice 1c used it, didn't add it.

Review verdict: branch ready to merge to main.

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

docs/ISSUES.md

@@ -46,6 +46,58 @@ Copy this block when adding a new issue:
 
 # Active issues
 
+## #60 — `obstruction_ethereal` retail downstream path not ported (M2 combat-HUD impact)
+
+**Status:** OPEN
+**Severity:** LOW for M1 (no observable defect); MEDIUM for M2 (combat contact reporting on ethereal creatures will be wrong)
+**Filed:** 2026-05-13 (final-review surfaced from B.4b)
+**Component:** physics / `CollisionExemption.ShouldSkip` + downstream movement contact handling
+
+**Description:** B.4b's L.2g slice 1b widened `CollisionExemption.ShouldSkip` to exempt
+on `ETHEREAL_PS` alone (cite `src/AcDream.Core/Physics/CollisionExemption.cs:62-79`). Retail's
+`acclient_2013_pseudo_c.txt:276782` requires both `ETHEREAL_PS && IGNORE_COLLISIONS_PS` to wrap
+the entire `FindObjCollisions` body — ETHEREAL alone takes the deeper path at line 276795 which
+sets `sphere_path.obstruction_ethereal = 1` and lets downstream movement allow passage WHILE
+STILL REPORTING THE CONTACT. We do not port that downstream path; we just exempt entirely.
+
+**M2 impact:** Combat HUD work that relies on physics-contact reporting for ethereal creatures
+(ghosts, partially-phased monsters, spell projectiles with ETHEREAL set) will see no contact at
+all instead of "soft contact with obstruction_ethereal=1". The user will not be able to target
+or interact with such entities via the contact path.
+
+**Acceptance:** Port the retail deeper path so `obstruction_ethereal=1` flows through movement +
+collision-reporting layers. Tests should cover: ETHEREAL creature target → contact reported but
+passage allowed; ETHEREAL+IGNORE_COLLISIONS target (door, retail-style) → full exempt.
+
+**Estimated scope:** Moderate. Touches `CollisionExemption.cs`, transition/movement layer, and
+sphere-path state propagation. Visible test through a spawned ethereal creature in ACE.
+
+---
+
+## #59 — `WorldPicker` 5m fixed-radius could over-pick at tight thresholds (M1-deferred polish)
+
+**Status:** OPEN
+**Severity:** LOW (cosmetic — picker grabs the right entity in Holtburg-tested scenarios)
+**Filed:** 2026-05-13 (final-review surfaced from B.4b)
+**Component:** selection / `AcDream.Core.Selection.WorldPicker.Pick`
+
+**Description:** `WorldPicker.Pick` uses a hardcoded 5m sphere around every candidate's
+`Position` regardless of the entity's actual size (`src/AcDream.Core/Selection/WorldPicker.cs:82`).
+This matches `WorldEntity.DefaultAabbRadius` and is sufficient for M1 acceptance: in tight
+doorways, every server-keyed candidate has correct sphere coverage and the closest-wins logic
+plus `ServerGuid==0` skip filter the wrong picks. But the invariant "non-clickable geometry has
+`ServerGuid==0`" is load-bearing — if L.2d ever ports `CBuildingObj` as a server-keyed entity,
+the picker may mis-target buildings. Per-entity `Setup.Radius` would be tighter.
+
+**Acceptance:** Either (a) tighten picker to read per-entity Setup.Radius / CylSphere bounds,
+or (b) document the invariant in `WorldPicker.cs` and add a regression test asserting
+`ServerGuid==0` entities never reach the per-candidate hit test.
+
+**Estimated scope:** Quick (~1 hour) — wire `Setup.Radius` lookup into the picker and update
+the 6 existing picker tests with realistic radii.
+
+---
+
 ## #58 — Door swing animation: UpdateMotion not wired for non-creature entities
 
 **Status:** OPEN

docs/research/2026-05-13-b4b-shipped-handoff.md

@@ -249,9 +249,9 @@ detection working, the correct gate firing, and `CollisionExemption`
 widened, the registry still held the stale closed state and the door
 stayed solid.
 
-**Fix:** Added a `_entitiesByServerGuid` reverse-lookup dictionary to
+**Fix:** Used the pre-existing `_entitiesByServerGuid` reverse-lookup
-`GameWindow` (populated at entity registration in `OnLiveCreateObject`).
+dictionary on `GameWindow` (populated at entity registration in
-`OnLiveStateUpdated` now does:
+`OnLiveCreateObject` since Phase 6.6/6.7). `OnLiveStateUpdated` now does:
 
 ```csharp
 if (_entitiesByServerGuid.TryGetValue(parsed.Guid, out var entity))