Every external check the build pipeline runs against an official figure — reform scores, tax-expenditure values, and baseline actuals — grouped into suites by what they test. Out-of-sample suites are the genuine fidelity tests; in-sample suites confirm the calibration converged. populace is a static calendar-year liability estimate on uprated survey data, so small gaps vs cash-receipts scores are expected. Trend and Δ track each check's |error| across releases (down and green is better).