# Documentary assembly workflow

The order matters. Most editors burn weeks because they do this in the
wrong sequence.

## Phase 0 — material in hand

- All cards ingested + verified (see `01-wedding-day-backup.md`)
- All interviews transcribed (Resolve Studio's built-in Transcribe is
  good now — Window → Subtitles → Create Subtitles from Audio)
- All B-roll logged in the proxy folder
- Music wishlist + at least 2 placeholder tracks already in `06_MUSIC_SFX/`

## Phase 1 — radio edit (text only)

Pick interview lines, ignore visuals.

1. Drop all transcript SRTs into `05_CLAUDE/transcripts/`
2. Open the Claude.ai Project for this film. Drag the transcripts in.
3. Ask: *"Build a radio edit targeting 8 minutes. Use BBC-Earth-style
   narrative arc — exposition / hero / villain / valley / crescendo /
   release. Output each selected line as `[speaker] [in: HH:MM:SS] [out:
   HH:MM:SS]`."*
4. Iterate with Claude until the **text** reads like a coherent story.
   Don't open Resolve yet.
5. Save final selection as `05_CLAUDE/cut_notes/radio-edit-v3.md`

## Phase 2 — outline → timeline

Convert the radio edit into a working Resolve timeline.

1. Ask Claude in the same Project: *"Give me the outline of that radio
   edit in the Bridge format with HH:MM:SS — BEAT NAME headings."*
2. Save as `05_CLAUDE/outlines/v1.md`
3. Bridge it:

   ```bash
   python3 ~/kit/bridge.py 05_CLAUDE/outlines/v1.md --fps 23.976 \
       --project "Architect Doc v1" \
       --out 04_RESOLVE/outline-v1.fcpxml
   ```

4. In Resolve: File → Import → Timeline → pick `outline-v1.fcpxml`.
   Every storytelling beat is now placed at the right timecode.

## Phase 3 — drop interview audio under the beats

For each placeholder beat:

1. Find the matching interview clip in the Media Pool
2. Drop it on A1 under the beat
3. Trim to the exact in/out from your radio edit

By the end of Phase 3, you should have audio-only telling the whole
story. **Watch the full thing with your eyes closed.** If it doesn't
hold up as audio, no visual save will fix it.

## Phase 4 — visual coverage pass

Now and only now, layer B-roll.

For each beat, ask:
- What single image carries the emotional weight here?
- What's the establishing wide?
- Where do we cut for movement?

Drop B-roll on V2 above the interview track. Resist the urge to colour
or grade yet.

## Phase 5 — picture lock

- Watch with someone outside the project. Note every place they get bored.
- Cut the boring bits ruthlessly.
- Lock the cut.
- Export `04_RESOLVE/picture-lock-v1.drp` as a safety snapshot.

## Phase 6 — music, then SFX, then grade, then mix

In this order. Music first defines emotional pacing — never grade before
music. Grade last because grading drives mood, and mood is finalised by
the music.

## Phase 7 — deliverables

Use the cutdowns CLI for social pieces from the master:

```bash
python3 ~/kit/cutdowns.py 07_EXPORTS/final/master.mov \
    05_CLAUDE/cut_notes/social.md --aspect 9:16 \
    --out 07_EXPORTS/social/
```

Renders normalised IG-ready cuts at -14 LUFS.

## Rules of thumb

- **If you're editing visuals before locking audio, you're wasting time.**
  Audio-locked first, always.
- **If you can't describe each beat in one sentence, the cut isn't ready.**
  Go back to the outline.
- **If a B-roll cut needs a heavy grade to work, it's the wrong shot.**
  Find a different one.