1d60f0da14
- F-01: cost bar crit (red) state past 90% of budget - F-02: normalize() hardens fetched board state against sparse payloads - F-03: /thumb 404s for videos with no extractable poster frame instead of serving raw video bytes - F-04: checkpoint artifact path refs only resolve inside the project dir - F-05 (board half): stall detection — in_progress stage with no disk activity >10min renders red 'stalled?' + header badge flips to STALLED?; verified against the real wedged why-cities-glow project - eval harness from dogfood session committed (visual regression + interaction smoke, capture watcher, server/gate test suites) + regression tests for each finding; 46 backlot tests green, visual eval green (restage-before-capture note logged)
Backlot — the living storyboard
A read-only local board that shows a production happening: pipeline stages
lighting up, the script as a screenplay page, the scene plan as a filmstrip
that fills in as assets generate, decisions, spend, and activity — all
derived from what the pipeline already writes to projects/<id>/.
python -m backlot open <project-id> # start server if needed + open browser
python -m backlot open # library view (all projects)
python -m backlot serve --port 4750 # run the server in the foreground
How it stays live
No agent involvement. A watchfiles watcher on projects/ publishes change
notifications over SSE; the browser refetches board state. State sources:
| Board element | Disk source |
|---|---|
| identity / rail order | project.json + pipeline_defs/<type>.yaml |
| stage states, gates, versions | checkpoint_<stage>.json + history/ |
| script card / modal | artifacts/script.json |
| filmstrip cards | scene_plan × script × asset_manifest join |
| generating shimmer, activity | events.jsonl (written by BaseTool instrumentation) |
| cost meter | checkpoint cost_snapshot |
| renders | renders/*.mp4 (+ root-level mp4 heuristic) |
Projects without checkpoints degrade gracefully to a "what the watcher found" view — media, snapshots, renders.
Replay: a completed run can be scrubbed end-to-end (▶ REPLAY RUN on the board) — reconstructed from checkpoint history and event timestamps.
Try it without a real production:
python scripts/backlot_simulate_run.py # live demo run (~1 min)
python -m backlot open backlot-demo-run
Design doc: internal/design/LIVING_STORYBOARD.md.