diff --git a/MEMORY.md b/MEMORY.md
index 7074fdd..028ae68 100644
--- a/MEMORY.md
+++ b/MEMORY.md
@@ -11,8 +11,12 @@ Session memory for durable decisions, state, and gotchas. Keep concise, remove o
   direct `set_bone_pose_rotation` on HipL/KneeL/Spine/ShoulderL/ElbowL/Head changes rendered
   pixels; a bone *position* offset moves the head; a standalone probe running the same
   `_animate()` sequence renders the crouch. A staged in-game `film` frame also showed the arms
-  down, so live venue animation may be affected too. Do not treat the pose sheets as evidence;
-  a scene test (posed frame differs from rest frame) is the next step.
+  Standalone SceneTree probes (avatar under `root`, under a `Node3D` stage, and created
+  mid-`_process`) all render the seated crouch, so the rig and animation math are fine and the
+  harness capture path is the outlier. A headless test now locks the proxy→bone contract
+  (`test_pose_reaches_the_skeleton_bones`, suite 273). Next leads: reproduce the stage inside
+  `capture_npc_review.gd` (known-good captures) or capture via `SubViewport` +
+  `RenderingServer.force_draw()`.
 - **3D review baseline (2026-11-03):** `docs/3d-model-review.md` records the first skill-guided
   measurements: 46/46 assets clean, library 59,012 tris, characters 4,482–5,474 tris (68–69% of
   the 8,000 caps), `npc_base` only 1,204/2,488, 24-NPC 6.94 ms, and authored texel density
diff --git a/TODO.md b/TODO.md
index d44b9cf..f3d81e8 100644
--- a/TODO.md
+++ b/TODO.md
@@ -64,10 +64,12 @@ except the rebuilt assets themselves.
       frames still show the rest pose; direct bone writes deform the mesh, so this is a staging/
       capture discrepancy that may also affect live venue frames. See
       `docs/3d-model-review.md` §"Pose evidence is blocked".
-- [ ] Diagnose the pose-capture discrepancy: compare the harness staging with the working
-      standalone probe (SceneTree lifetime, node parents, first-frame skin bind, capture timing)
-      and add a scene test asserting a posed avatar renders differently from its rest frame.
-      Do not trust pose sheets or claim rig deformation until that test passes.
+- [ ] Finish the pose-capture diagnosis. Done so far: the pose reaches proxies and skeleton
+      bones (new headless test, 273/273), direct bone rotations deform the rendered mesh, and
+      three standalone probes render the seated crouch — only `capture_npc_pose.gd` saves the
+      rest pose. Remaining leads: reproduce the stage inside `capture_npc_review.gd` (known to
+      render) or capture via `SubViewport` + `RenderingServer.force_draw()`, then add a scene
+      test that a posed avatar renders differently from its rest frame.
 - [ ] UV/texel density: measure the 256² slot maps; decide whether atlas, UDIM or a resolution
       change is justified. Do not change `CHAR_MAX_TRIS`/`CHAR_MAX_VERTS` without re-running the
       budget smoke and the Web/Compatibility path.
diff --git a/docs/3d-model-review.md b/docs/3d-model-review.md
index 1a370ba..c23aeea 100644
--- a/docs/3d-model-review.md
+++ b/docs/3d-model-review.md
@@ -165,26 +165,27 @@ for presence, not re-reviewed for art quality.
 
 ## Pose evidence is blocked (2026-11-03)
 
-The weights/pose column cannot be filled yet. A pose harness was added
-(`src/test/capture_npc_pose.gd`, staged in `tmp/model-pose*`), and its output does not match the
-state it reports:
+The weights/pose column cannot be filled yet. The pose logic is verified; the *capture* is not.
+`src/test/capture_npc_pose.gd` stages idle / walk-contact / walk-pass / seated for all eight
+variants (full-body + shoulder views) and its output does not match the state it reports.
 
-- `_animate()` applies the seated pose (logged `rig_y=-0.339`, `leg0=-1.357`) and
-  `Skeleton3D.get_bone_pose_rotation()` returns the rotated quaternion at capture time, yet the
-  saved frame shows the rest pose.
-- Ruled out: missing skeleton (one `Skeleton3D`, mesh bound to it, `same=true`), hidden mesh
-  (hiding `npc_base` removes the body), pose not applied (proxies and bone poses verified),
-  and unbound skinning: writing a single bone rotation directly (`HipL`, `KneeL`, `Spine`,
-  `ShoulderL`, `ElbowL`, `Head`) changes the rendered pixels, and a bone *position* offset moves
-  the head. A standalone probe that calls the same `_animate()` sequence renders the crouch.
-- Also observed: a staged in-game `film` pose frame showed the arms down, so this may affect the
-  live game rather than only the harness. That is why it is a P0/P1 candidate rather than a QA
-  tooling nit.
+| Check | Result |
+| --- | --- |
+| Pose reaches proxies | Seated pose logs `rig_y=-0.339`, `leg0=-1.357` |
+| Pose reaches skeleton bones | `Skeleton3D.get_bone_pose_rotation(HipL/KneeL)` differs from rest; new headless test `test_pose_reaches_the_skeleton_bones` locks it (suite 273/273) |
+| Idle restores rest | Same test asserts the hip returns to its rest orientation |
+| Skinning bound | One `Skeleton3D`, mesh bound to it (`same=true`); hiding `npc_base` removes the body |
+| Bone rotations deform the mesh | Direct writes change rendered pixels: Spine 1,719 / HipL 1,052 / KneeL 810 / ElbowL 745 / Head 671 / ShoulderL 640 sampled pixels |
+| Bone translation deforms the mesh | A +0.5 m Head offset moves the head in frame |
+| Standalone probes render the pose | Three variants (avatar under `root`, under a `Node3D` stage, created mid-`_process`) all render the seated crouch (`posed` vs rest diff ≈ 2,200–2,360 samples) |
+| The harness renders the pose | **No.** All 8 variants × 4 poses save the rest pose while the logged state and bone poses are posed |
 
-Next step is a focused diagnosis: compare the harness/venue staging with the standalone probe
-(SceneTree lifetime, node parents, first-frame skin bind, capture timing) and add a scene test
-that asserts a posed avatar renders differently from its rest frame. Do not mark the pose/weight
-scorecard row until that test exists.
+The cause is therefore in the harness's capture path, not in the rig or the animation math. Two
+leads are left and both are cheap: reproduce the stage inside a harness that is known to render
+correctly (`capture_npc_review.gd` already produces 96 valid frames), or capture through a
+`SubViewport` with `RenderingServer.force_draw()` so the frame cannot lag the state. Until a
+scene test asserts that a posed avatar renders differently from its rest frame, do not treat pose
+sheets as evidence and do not claim rig deformation in the review.
 
 ## Ranked candidate gaps
 
diff --git a/src/test/capture_npc_pose.gd b/src/test/capture_npc_pose.gd
index c804b30..a9e1c59 100644
--- a/src/test/capture_npc_pose.gd
+++ b/src/test/capture_npc_pose.gd
@@ -261,8 +261,6 @@ func _shoot() -> void:
 	while FileAccess.file_exists(path):
 		path = dir.path_join("%s-v%d.png" % [base, index])
 		index += 1
-	var hip_bone: int = int(_avatar._bone_indices.get("HipL", -1))
-	var knee_bone: int = int(_avatar._bone_indices.get("KneeL", -1))
 	var img: Image = root.get_texture().get_image()
 	var err := img.save_png(path)
 	print("%s -> err=%d" % [base, err])
diff --git a/src/test/tests/test_npc_character.gd b/src/test/tests/test_npc_character.gd
index d93dbe0..47b4876 100644
--- a/src/test/tests/test_npc_character.gd
+++ b/src/test/tests/test_npc_character.gd
@@ -147,6 +147,38 @@ func test_skinned_model_loads_and_keeps_the_head_contract() -> bool:
 	avatar.free()
 	return a and b and c
 
+## The pose harness captures frames, but a regression here is cheap to catch:
+## the pose logic must reach the skeleton bones, not only the proxy pivots.
+func test_pose_reaches_the_skeleton_bones() -> bool:
+	var avatar := _avatar(Rules.NpcType.TOURIST, 55)
+	var sk := avatar._skeleton
+	var hip: int = int(avatar._bone_indices["HipL"])
+	var knee: int = int(avatar._bone_indices["KneeL"])
+	var rest_hip := sk.get_bone_rest(hip).basis.get_rotation_quaternion()
+	var rest_knee := sk.get_bone_rest(knee).basis.get_rotation_quaternion()
+	avatar.arrived = true
+	avatar.npc.intent = "sit"
+	for _i in 45:
+		avatar._animate(1.0 / 60.0)
+	var sit_hip := sk.get_bone_pose_rotation(hip)
+	var sit_knee := sk.get_bone_pose_rotation(knee)
+	var a := TestUtils.check(not sit_hip.is_equal_approx(rest_hip),
+		"seated pose rotates the HipL bone away from rest")
+	var b := TestUtils.check(not sit_knee.is_equal_approx(rest_knee),
+		"seated pose rotates the KneeL bone away from rest")
+	var c := TestUtils.check(avatar._rig.position.y < -0.25,
+		"seated pose lowers the rig (%.3f)" % avatar._rig.position.y)
+	# Returning to idle must restore the rest orientation, not leave the sit pose.
+	avatar.arrived = false
+	avatar.npc.intent = "roam"
+	avatar.velocity = Vector3.ZERO
+	for _i in 120:
+		avatar._animate(1.0 / 60.0)
+	var d := TestUtils.check(sk.get_bone_pose_rotation(hip).is_equal_approx(rest_hip),
+		"idle restores the HipL rest orientation")
+	avatar.free()
+	return a and b and c and d
+
 func test_character_selection_is_deterministic_and_complete() -> bool:
 	var expected := {
 		Rules.NpcType.TOURIST: "npc_casual",