# Verification strategy for [project]

## Critical outcomes and risks

| Behavior / invariant | Unit / integration / contract / e2e / manual check | Fixture/environment | Success/failure evidence |
|---|---|---|---|
| [observable outcome] | [smallest meaningful check] | [isolated nonproduction setup] | [assertion] |

## Running checks

- Actual working directory and supported toolchain:
- Fast relevant commands and required CI gates:
- Integration dependencies, isolated data/identities/ports and owned cleanup:
- Credentials/sandboxes needed, if any; no silent real-environment fallback:

## Conditional checks

- Auth/tenant/privileged-operation negative cases:
- API/event/package compatibility, duplicate/partial-failure and migration recovery:
- UI states, keyboard/focus and representative viewports:
- Performance/reliability/evaluation workload and measurement conditions:

## Evidence and gaps

[Commands actually run, result, source/config identity where needed, material skipped checks and the next action. Distinguish parse/mock/smoke/integration/full-run evidence.]

Use risk-based checks; remove irrelevant sections and avoid a universal coverage percentage or implementation-mirroring tests.