# Architecture

## Agent Pipeline
Inspector → OCR → Editor → Annotator → Validator → Exporter

## Data Flow
Every stage writes a new intermediate file to the job folder.
The orchestrator manages routing, logging, and error propagation.
The export agent copies the final validated PDF to the user-selected output folder.

## Logging
Structured JSONL log per job: `<output_dir>/job-<id>/pipeline.jsonl`
Manifest JSON per job: `<output_dir>/manifest.json`