Frame Skeleton

You classify and lay out the spine only. You do not pick templates, write copy, choose colors/fonts, or decide a frame’s groups — those are Step 3 (the plan fills each frame in place). Leave every frame’s ### Groups as TBD (Step 3) and the frontmatter style blank.

There is no intermediate JSON — the skeleton is the start of STORYBOARD.md. Step 3 edits the same file.

The trust boundary (read this first)

audiomap.json is one analyzer’s output. Some fields are robust on any music; some are reliable only when the music is actually rhythmic. This decides each frame’s pacing:

• Grid is reliable when: rolls present, and/or dense phases, and/or high onset_rate with a steady grid.
• Grid is fictional when: rolls≈0, mostly sparse phases, low onset_rate → pace by phrases[] + energy_phases[], not beats.

How to lay out frames (run in order)

1. Gestalt. From summary / tempo / audio.duration_sec + the roll count, write the frontmatter compositionId, duration_s (== audio.duration_sec), canvas, and a one-line read of the track’s density arc.
2. Cut into frames. Walk energy_phases[] and split where the music genuinely changes state — at hard_stops[], SURGE / DROP key_moments, the start/end of a rolls[] run, an onset desert (a long gap in events[]), or a big energy-level jump. Collapse adjacent phases that are one gesture. Expect ~1–6 frames; a short clip may be one. Snap every boundary to an audiomap anchor, then re-snap to the nearest beats_sec (tolerance ≤ ½ beat) only when the grid is reliable; on calm material snap to phrases[] / energy_phases[] edges instead. Frames tile the track (first at 0, last at duration_s, no gaps/overlaps).
3. Per frame, set pacing — the trust-boundary call:
   • beat_cut — genuinely rhythmic: a roll present, or dense, or (clearly high onset_rate and a steady grid). Hard cuts / per-onset reveals may anchor to beats.
   • phrase_flow — calm / sparse: rolls≈0, mostly sparse, low onset_rate. Do not anchor hard cuts to the grid; pace by phrases[] + the energy envelope (slow crossfades, long holds).
4. Per frame, tag mood (1–3 of: warm · dark · hype · elegant · glitch · cinematic · playful · tense · dreamy · aggressive) from energy_phases[].feel + energy + any genre cue in the brief/title.
5. Per frame, write a one-line feel — the plain-language music situation Step 3 matches a template against (e.g. “accelerating onset stream into a held downbeat”, “calm held pad, one onset desert”, “fast sustained-fill roll, no readable message”). This is what the planner reads against the catalog’s Reach for it when — keep it concrete, drawn from the robust fields, never invented.

What the skeleton looks like

A valid STORYBOARD.md with the spine set and every frame’s treatment left for Step 3 (full syntax in storyboard-format.md):

Self-check

• duration_s == audiomap.audio.duration_sec; frames tile the track gap-free (first at 0, last at duration_s).
• Every frame has src + span_sec + duration + pacing + mood + a one-line feel.
• pacing was set from the robust fields (energy / density / rolls / onset_rate), never from bpm / beats_sec alone.
• No frame boundary sits inside a rolls[] run or leaves a sub-1-bar fragment.
• Every frame’s ### Groups is TBD (Step 3); style is blank. No template, copy, color, or font anywhere.