rules.example.yaml

# ingestlib content rules — copy to rules.yaml beside config.yaml. Optional:
# without it, classify is open-ended and split discovers its own vocabulary.
# Per-call arguments to ingest()/classify() always override this preset.

classify:
  max_pages: 5                 # cap pages read (after target_pages); omit = no cap
  target_pages: "1,3,5-7"      # 1-based pages/ranges (omit = all); usually per-document — override per call
  rules:                       # up to 20 — result is one of these or "uncategorized"
    invoice: "Itemized charges, tax info, and payment terms"
    sec_filing:                # extended form → auto-extract on ingest, stored in the registry
      description: "10-K/10-Q style regulatory filings"
      auto_extract: true
      extract_schema:          # JSON Schema — the shape to pull out
        type: object
        properties:
          fiscal_year:
            type: integer
          total_revenue:
            type: number
        required: [fiscal_year]

split:                         # YOUR section vocabulary instead of LLM-discovered
  unmatched: other             # leftover pages: other=group (default) | require=force each | skip=drop
  categories:                  # up to 50 — {section_name: description}
    financial_statements: "Balance sheets, income statements, cash flow statements"
    notes: "Footnotes and accounting-policy disclosures"