Job schema and database

Job states

State	Meaning
queued	Waiting in the worker
running	Processing in progress
completed	Transcription (and summary if requested) finished
failed	Irreversible pipeline error
cancelled	Cancelled by user (only from queued)

Phases (phase)

During running, the pipeline updates phase and progress_pct:

Phase	Description
idle	Initial
extract	Audio extraction from video
transcribe	NeMo ASR
summarize	Summary
export	Writing final files
done	Completed

SQLite table jobs

File: data/output/jobs/queue.db

Column	Type	Description
id	TEXT PK	YYYYMMDD_HHMMSS_stem
stem	TEXT	Filename without extension
source_name	TEXT	Original filename
created_at	TEXT	ISO creation timestamp
output_dir	TEXT	Absolute job folder path
input_path	TEXT	Source copy path in job
status	TEXT	Current state
phase	TEXT	Pipeline phase
progress_pct	REAL	0–100
progress_message	TEXT	UI message
queued_at	TEXT	Enqueued
started_at	TEXT	Processing started
finished_at	TEXT	Finished (success or error)
error	TEXT	Pipeline error
has_summary	INTEGER	1 if summary saved
summary_requested	INTEGER	1 if requested
summary_error	TEXT	Summary-only error
transcript_chars	INTEGER	Text length
model_name	TEXT	NeMo model used
device	TEXT	cpu / cuda
summary_mode	TEXT	extractive / abstractive
summary_length	TEXT	auto / short / normal / detailed

Indexes: idx_jobs_status, idx_jobs_created.

Files per job

Folder: data/output/jobs/YYYYMMDD_HHMMSS_stem/

File	Content
source.*	Copy of uploaded file
job.json	JSON mirror of SQLite record
trascrizione.txt	Full text
sottotitoli.srt	Subtitles with timestamps
riassunto.txt	Summary (if generated)
work/	Temp files (WAV, chunks) — may remain

Job ID

Format: {timestamp}_{stem_sanitized}

Example: 20260628_143022_campione-italiano-lungo

The stem is truncated and sanitized of unsafe filesystem characters.

Migration

On startup, if the old index.json exists and queue.db is empty, records are imported automatically.