document_processing

Document Processing Workflow

Pattern: Fan-out / merge (5 steps) Demonstrates: Parallel step execution, multi-input merge, DAG scheduling

Use case

Your team ingests 200+ documents per day — contracts, invoices, reports — and needs each one classified, entity-extracted, and summarized before it enters a review queue. This workflow processes a document in parallel across three specialized steps (classify, extract, summarize), then merges the results into a single structured record.

The fan-out pattern means the three parallel steps run concurrently after ingestion, with the merge step waiting for all three to complete before producing the final output.

Steps

ingest ──┬──► classify ──┐
         ├──► extract  ──┼──► merge
         └──► summarize ─┘

Step	Role	Capability
ingest	Load document from storage	fs.read (live mode)
classify	Determine document type and priority	llm.call (live mode)
extract	Extract named entities (parties, dates, amounts)	llm.call (live mode)
summarize	Generate a concise summary	llm.call (live mode)
merge	Combine all results into a structured record	(pure)

How to run

# Validate the workflow DAG
cargo run --bin boruna -- workflow validate examples/workflows/document_processing

# Run in demo mode
cargo run --bin boruna -- workflow run examples/workflows/document_processing --policy allow-all

# Run and record evidence
cargo run --bin boruna -- workflow run examples/workflows/document_processing --policy allow-all --record

Evidence produced

Each --record run writes a bundle to evidence/run-document-processing-<timestamp>/ containing:

• audit_log.json — hash-chained log of all 5 steps (includes parallel step entries)
• policy.json — policy snapshot
• outputs/ — per-step output values (all 5 steps)
• env.json — environment fingerprint

Notes

• The classify, extract, and summarize steps are independent and scheduled concurrently by the workflow runner.
• The merge step declares depends_on: ["classify", "extract", "summarize"] in workflow.json to enforce the merge barrier.
• In demo mode, all three parallel steps return realistic hardcoded outputs for the sample contract document.