What is 'the move' that has quietly defined the last 20 years of serious industrial programming?	Pull a property from 'things we hope to catch at runtime' into 'things the compiler refuses to build without.' TypeScript, Rust, TLA+, Zelkova, OPA, differential privacy — all the same move, in different domains.
Why is a reconciliation job a 'runtime confession of a missing compile-time guarantee'?	A reconciliation is a nightly cross-check between two independently-computed views of the same truth, run because you cannot prove they agree. If the two views were proved equivalent by construction, the reconciliation would have nothing to do.
What is 'grain' in data modelling, and why has Kimball's 2003 naming of the problem not fixed it?	Grain is the combination of columns at which records in a table are unique — a binding contract on a data model. Kimball called it 'the most frequent design error by far' in 2003. The state of the art in 2026 is still 'think deeply and write a uniqueness test' because no mainstream tool makes grain a language-level property the compiler can refuse to violate.
Why is data engineering the right first vertical for a guarantee-making system — rather than, say, OS kernels or crypto?	Three reasons: (1) the gap between the guarantees practitioners want and what their tools provide is widest there; (2) the diagnoses are already written — Kimball on grain, Codd on relational closure, Hellerstein on declarative dataflow, Brandon on SQL; (3) the cost of the status quo is both well-quantified (~2 days/week firefighting, ~$2T/year software quality) and paid in cash, monthly, by every company that ships dashboards.