← All research

The core theories / Saves compute

Affected Verification

Do not verify what a change cannot affect.

Choose verification from defensible change-impact evidence. When the boundary is unknown, broaden the checks or stop.

Explore the repository

The question

Which checks can this change affect?

Affected Verification plans the smallest verification workload it can defend using change impact, dependencies, coverage, policy, and risk evidence. It explains every selected or skipped check. The consuming application owns execution.

In practice

A concrete example.

A change outside a test’s complete dependency boundary can justify skipping that test. An incomplete boundary cannot. Unknown impact broadens verification instead of silently becoming a pass.

Evidence & limitations

What the implementation supports.

Research remains narrow. A Python shadow experiment recorded a missed check; a later repair selected that known check. This does not prove general selector completeness. Targeted selection is defensible only with complete manifest-backed evidence.

Last verified 2026-11-16. 126 tests pass and 1 is a documented skip. Determinism and conformance pass.

Read the experiments, repair, and claim limits ↗

Independent by design.

Explore this idea on its own, or see how the four theories fit together.

See the bigger picture