Import runs

Every import extracts facts with provenance, proposes changes, and waits for review. Nothing writes to the catalog without an approval trail — except unambiguous updates, which auto-apply and stay auditable.

How a run works
  1. Extract — parsers pull atomic facts (price, header, suffix, option mention) from each source with a locator and a confidence score.
  2. Compare — facts are diffed against the current catalog and against each other; agreeing sources raise confidence, disagreeing sources create conflicts.
  3. Propose — changes become proposals (price update, new item, discontinue, new rule, new resolver). Unambiguous ones (single source, matches expectations, high confidence) auto-apply.
  4. Review — everything else waits in the run's proposal list or the review queue, with a preview of exactly which items change and how.
  5. Watch — applied changes that affect finalized/sent quotes flag those quotes for attention; the frozen documents never mutate.