rules.example.yaml

# ingestlib content rules — copy to rules.yaml beside your config.yaml.
#
# Infra lives in config.yaml; what your documents MEAN lives here. The file
# is optional: without it, classification is open-ended and split discovers
# its own section vocabulary. Per-call arguments always override this preset.

# Classification preset — used by classify() and the ingest pipeline
# whenever no explicit arguments are given.
classify:
  max_pages: 5                 # cap on pages read (after target_pages); omit for no cap
  target_pages: "1,3,5-7"      # 1-based pages/ranges to read; omit for all pages
  rules:                       # up to 20 — result is one of these or "uncategorized"
    invoice: "Itemized charges, tax info, and payment terms"
    sec_filing: "10-K/10-Q style regulatory filings"

# Split preset — YOUR section vocabulary instead of the LLM-discovered one.
# Pages fitting no category follow `unmatched`:
#   other (default) — grouped into an honest `other` section
#   require         — every page forced into a category (left-neighbor repair)
#   skip            — dropped entirely (no sections, no chunks, no vectors)
split:
  unmatched: other
  categories:                  # up to 50 — {section_name: description}
    financial_statements: "Balance sheets, income statements, cash flow statements"
    notes: "Footnotes and accounting-policy disclosures"