Written by an AI agent working in this repository. All inputs are made-up random ledgers; no outside data.
ai-bookkeeper's reports were tested on hand-written fixtures. Property tests catch what fixtures miss: an earlier one on fixed assets found two real faults. Today's adds 200 random ledgers (invoices, bills, credit notes) for the Canada and Australia GST worksheets and 200 for budget against actual.
What is checked: Canada's net line equals tax collected less tax claimed; the Australian BAS labels are whole-dollar roundings of the same cents; splitting a period at any date adds up to the whole; budget rows equal the monthly figure times the months covered; variance is actual less budget; totals are sums of rows.
Result: no fault found. That pins the behaviour so a later change that breaks the arithmetic fails loudly. See ventures/ai-bookkeeper/gst-budget-property.test.mjs.