Storage and persistence

Data layout

data/                          ← SBOBINATOR_DATA (default: project_root/data)
├── input/                     ← source files (manual or upload)
│   └── README.md
└── output/
    ├── jobs/
    │   ├── queue.db         ← SQLite WAL
    │   ├── worker.pid       ← active worker PID
    │   └── YYYYMMDD_HHMMSS_name/
    │       ├── source.wav
    │       ├── job.json
    │       ├── trascrizione.txt
    │       ├── sottotitoli.srt
    │       └── riassunto.txt
    └── benchmark_*.json/md    ← optional benchmark reports

models/                        ← NEMO_CACHE_DIR
├── parakeet-tdt-0.6b-v3.nemo
└── mt5-small/
    ├── config.json
    ├── model.safetensors
    └── spiece.model

SQLite queue.db

• jobs table with all JobRecord fields
• WAL mode for concurrent reads (UI poll + worker writes)
• Automatic migration from legacy index.json (if present)

What happens if...

Action	UI list	Text files
App restart	✅ Persists	✅ Persists
Delete job folder	⚠️ Job in DB, files missing	❌
Delete only queue.db	❌ Empty list	✅ Files remain
clean_output.py	❌ Empty	❌ Everything removed

job.json

JSON copy of the SQLite record, synced on every update. Useful for manual inspection and backup; the UI does not use it for the list.

Git

.gitignore excludes:

• data/output/* (except .gitkeep)
• data/output/jobs/* (except .gitkeep)
• models/* (except .gitkeep)
• queue.db

Results must not be committed to git.

For the full schema see Job schema and database.

Docker volume

Only ../data:/data is mounted from the host. Models are in the image (/models), not in the volume.