Windowed vs continuous

Windowed vs continuous - the two shapes a streaming result can take, and the windowed vs continuous query choice decides everything downstream: either one finished snapshot per bucket of time, or one running answer per key that never finishes. A windowed query emits each group once, when its window closes; the row is final and the output is append-only. A continuous query keeps one live value per key and republishes it as it changes, so the output is a changelog and the current answer is latest-per-key. That second shape is a materialised view, kept fresh.

Choose by the question. "Revenue per hour" is windowed; "revenue so far" is continuous. The sink follows the choice: an append-only store for finished windows, an upsert-capable store or a compacted topic for a changelog.

Do you need it? You need to make the choice, once, and it is one line of SQL. In ubik a GROUP BY with a tumbling window emits each window once when the watermark passes its end; the same GROUP BY without a window is a continuous view, republished as an upsert changelog on the --emit-every cadence. Same aggregates, same stream, different clock on the answer.