Choreo Phrase Authoring

Choreo Phrase Authoring

Use this when the problem is the dance, not just one bad numeric key. The goal is to turn movement feedback into labelled, reusable choreography data.

Read First

• lib/features/character/README.md
• lib/features/character/docs/adr/CHAR-0001-dance-choreography-encoding-and-move-library.md
• lib/features/character/docs/research/2026-06-28-afrobeats-dance-moves.md
• lib/features/character/model/dance_phrase.dart
• lib/features/character/model/afrobeats_move.dart
• lib/features/character/samples/cat_in_suit.dart

Authoring Principles

• Choreography is data. Add or adjust DancePhrase, AfrobeatsMove, DanceMoveSignature, DanceRoleStyle, accents, supports, and IK arcs before scattering anonymous per-bone key tweaks.
• Label the move and the count. A future reviewer should be able to say "frame 16 Buga hit" or "slot 2 backup Azonto answer" and find the data.
• Separate choreo from camera. Use static-camera review while building the move; re-enable camera only after the body motion works.
• Keep support physics plausible. Hip/root movement must match the feet; dancers should not drift forward/back in ways their legwork cannot support.
• Avoid clone sync. The cats can share a phrase, but style overlays should add bounded variance in timing, amplitude, expression, gaze, and role.
• Use dynamics, not only positions. DanceDynamics controls anticipation, snap/sustain, and overshoot; microFrames/swingFrames place motion in the pocket.
• Keep faces happy unless the user asks otherwise. For the current dance, skip sad/angry expression cycling.

Workflow

1. Define the music grid. Use beat/downbeat data from dance-track-prep when available. Otherwise state the assumed BPM and phrase length.

2. Choose the move vocabulary. Pick named moves from the ADR/research notes or add a new AfrobeatsMove entry with:

   • name and origin/reference;
   • feel (onBeat, offBeat, halfTime);
   • featured region for camera decisions;
   • default DanceDynamics;
   • default sub-frame swing/pocket.

3. Lay out the phrase. Define slots/counts before touching bones. Decide which cat leads, answers, or supports in each slot.

4. Author support and body first. Root, hips, pelvis, chest, and support-foot spans sell the groove. Limbs decorate the weight transfer; they do not replace it.

5. Add hands/feet as IK target arcs. Prefer semantic arcs tied to torso, pelvis, or named cue frames over raw FK if the hand/foot is meant to hit a visible place.

6. Add role variance. Use DanceRoleStyle and target/joint/body accent overlays so backup cats differ without forking the whole routine.

7. Render and review. First use static camera. Then run character-motion-review-panel; if snaps are reported, run temporal-animation-diff.

Targeted Checks

Run the model tests for choreo data you touched:

fvm flutter test test/features/character/model/dance_phrase_test.dart \
  test/features/character/model/afrobeats_move_test.dart \
  test/features/character/model/dance_dynamics_test.dart

For runtime motion regressions, add only the focused files needed:

fvm flutter test test/features/character/runtime/temporal_motion_analyzer_test.dart \
  test/features/character/runtime/dance_beat_alignment_test.dart

Smell List

• The panel keeps saying "generic boy band": the move vocabulary is too vague or the hips/feet are not leading the groove.
• It feels robotic: every cat peaks on the same frames with the same amplitude.
• It feels snappy/jumpy: inspect dense temporal diffs before adding more keys.
• The upper body floats off the hips: root/hip/chest channels are fighting or support spans do not match visible footwork.
• Arms hide behind the belly: z-order, IK target depth, or silhouette staging is wrong; fix the path and draw order, not just amplitude.
• Camera hides the problem: disable camera movement while comparing frames.

See Also

• dance-track-prep for song beat/downbeat and lyric timing data.
• dance-lipsync for real mouth-shape cue generation.
• character-motion-review-panel for scored expert feedback.
• temporal-animation-diff for frame-by-frame discontinuity analysis.