Confirm the brand and manufacturer part number for each product. Values from the store are shown when available. Do not substitute a store SKU or guess missing details; use categorization when they are unavailable.

Catalog research needs verified product identity. Missing brand/MPN values are held; store SKUs are not guessed to be part numbers. Workbook research is reused with no additional AI research.

The server selects from the imported catalog. Nonempty filter groups must all match; any value within a group may match. Leaving every filter empty includes all eligible simple products, subject to limits. Removed and unsupported products are excluded.

Use category IDs and brand values from this store's imported records. Brand filters use the actual catalog brand, not the saved research identifier override. Empty groups do not restrict selection.

Eligible product type: simple. Variants and media remain held until their complete workflow is supported.

Updates preserve price, inventory, options, and existing visibility. New products use explicit merchant commercial values from their workbook. Any warning, missing required value, conflict, or uncertain delivery is held.

Intervals range from 15 minutes to 7 days. AI limits cannot exceed USD 10,000 per run or USD 100,000 per UTC day. API costs are recorded from returned usage, without reserving money. These spending thresholds stop new requests after recorded spend reaches the limit; in-flight requests can exceed it. Missing costs remain unknown. Completed workbook research has already incurred its AI cost in Agents.