Batch Generation

For evaluation data you want volume and diversity. A batch turns one high-level brief into N distinct scenarios, then generates a document for each — every one paired with its ground-truth JSON label.

How a Batch Works

1. Scenario planning — a planner model reads your brief plus the schema's generation_guidance.md and produces N distinct, specific scenarios.
2. Concurrent fan-out — each scenario runs its own self-contained pipeline graph as a sibling node; Strands executes them concurrently.
3. Output — clean PDFs, data JSON, and optional augmented versions, each returned as a typed GeneratedDoc.

The planner varies entities, industries, regions, time periods, counts, and amounts so the set is genuinely diverse. A specific brief produces far more varied output than a generic one.

What to Expect

• A single document takes roughly 30 to 100 seconds end-to-end.
• All workers launch concurrently; the Python Strands graph has no built-in concurrency cap, so large counts may hit Bedrock rate limits (adaptive retries absorb this, but wall-clock grows). Start with modest counts.
• Partial failures are normal. Failed documents come back with success=False and verdict="error"; the rest still complete.
• Use a specific brief. Vague briefs produce same-y output.

Output Structure

output/
├── pdfs/                    # clean PDFs
├── data/                    # ground-truth JSON per document
└── generation_scripts/      # HTML files used to render the PDFs

Usage

• CLI Usage → Batch: --count > 1, the diversity brief, and batch flags.
• Python API Usage → Batch: Generator.generate_batch(...), on_document progress, seed, and the BatchResult.

Related

• Creating a Document Type: define the schema you batch over.
• Packets: coordinated multi-document sets.
• Models: choose models per stage.