Fewer Tests, More Proof

The unit of value in a test suite is a promise proven, not a test counted. A suite optimized under that metric gets smaller and stronger at once: each promise tested exactly once, with its strongest assertion, in every implementation and state where it applies — and nothing else. "Fewer tests" is the side effect, never the goal; the moment a consolidation would trade proof for count, it stops.

Inventory: map tests to promises

Before moving anything, build the map. For each test, name the promise it proves — the sentence that would have to become false for it to fail ("reusing a key with a different payload is refused as a conflict", "shutdown drains pending work"). A test whose promise you cannot name is the first finding. The map exposes the six consolidation targets:

1. The same promise tested per-implementation in N parallel files — the battery target.
2. The same setup rebuilt across many files — the shared-harness target.
3. A promise tested weakly in one place and strongly in another — the subsumption target.
4. A family of hand-picked examples all pinning one general promise at different points — the property target.
5. Tests that cannot fail for a nameable reason — assertion-free runs, "did not raise", tautological mocks. Ritual, not proof.
6. Probabilistic concurrency tests — sleeps, retries, run-it-100-times loops — the determinism target.

The six moves