We fed the AI bookkeeper's built-in keyword layer about 90 everyday descriptions (insurance, telecom, SaaS, ads, travel, fees, fintech) and counted the blanks. Fifteen got no suggestion.
Most are blank on purpose: IRS and EFTPS debits (payroll, estimated and sales tax look alike), health insurers and brokerages (usually an owner draw, not a business cost), a bare check number and a client's Zelle and a bare Amex payment. The rest were real gaps: tax-prep software, VPN plans and Google storage. Rules now cover them.
Caveat: the descriptions are made up from public company names, not a real export, so this finds gaps, not accuracy. A hunt for a third real labelled set found none we could reach; the two we have are scored in the README.