One governed running take, now reviewable end to end
A dedicated character-performance surface for video solve, DAZ retarget, declarative constraints, synchronized review, and governed delivery.
A01–A08 cache reuse and fail-closed routing
A cache hit saves inference cost; it never grants production admission or bypasses camera evidence.
A01–A08 source / DAZ frame-locked review
These are real cached outputs: source MP4, DAZ review MP4, FBX-derived Motion Clip and portable Motion Bundle. A valid bundle does not override a failed QA gate.
Review source intent against the governed DAZ take
Frame-locked A/B remains canonical evidence. Real-time 3D adds a non-destructive inspection orbit over the same synchronized FBX.
0:00.00 / 0:10.00single-stream lockP2-A machine gates and the hash-bound human Golden review are admitted
P2-A inherits the admitted P1-J body result, excludes low-confidence hand flips, estimates one take-level palm offset, and writes only lHand. Alignment 7/7, exported-FBX world-space 7/7, P2 detail 9/9, and the new 36-view Golden review pass. P1-J remains the production rollback.
240F / 24 FPS / 10.000s · Δ 0F
- Hip/knee phase
- 0.995
- Leading-leg match
- 100.0%
GVHMR incam supplies the take-level camera orientation directly; the source stays unmirrored and anatomical sides are never swapped.
- Body-yaw MAE
- 12.1°
- Torso MAE
- 7.5°
- Head / neck max Δ
- 4.4° / 3.7°
- Heading max Δ
- 7.0°
The immutable P0 contains no corrective writes. P2-A adds one fixed 22.35° world-X calibration on lHand only; per-frame observation following, finger, forearm, torso, leg, translation, and scale writes remain forbidden.
- Knee MAE L / R
- 7.6° / 6.5°
- Palm median L / R
- 23.5° / 21.7°
- Sole P95 L / R
- 27.7° / 29.0°
- F000 / first 5
- 0.214 / 0.205
P2-A passes target surface strain, frozen-P0 hip continuity, and zero exported-mesh penetration. The signed 36-view Golden receipt binds the exact FBX and GLB hashes, so P2-A is now open for production; P1-J remains the rollback.
- Lateral lean P95
- 6.1°
- Arm twist P95
- 16.6°
- Torso strain P95
- 1.28×
- Penetration frames
- 0 / 240
- Production
- YES
The left-palm agreement within 40° rises to 90.3% while head stability, F000/first-five correspondence, both soles, arm-torso clearance, and zero mesh penetration remain passing. P2-A is now admitted by the hash-bound human Golden review; P1-J remains the explicit rollback.
- Head jitter P95
- 2.5°
- F000 / first 5
- 0.214 / 0.205
- Palm P95 L / R
- 50.7° / 44.0°
- Sole P95 L / R
- 27.7° / 29.0°
- Min clearance L / R
- 0.43 / 0.45 SW