What problem does it solve?
Testing doctrine that helps you write tests (including agent-generated tests) that fail for the right reason, avoid common brittleness and flakiness traps, and ensure CI gates real regressions instead of producing decorative green.
Core Features & Use Cases
- Behavior-first test design: Uses practical invariants, clear boundaries, and resilient selector/assertion strategies to ensure tests protect what users actually experience.
- Flaky/fragile suite prevention: Provides a taxonomy of flaky test causes (timing, order/state pollution, concurrency, nondeterminism) plus workflows like quarantine with owners and fix-by dates.
- Agent safety gates for test generation: Enforces seven gates (invariant first, owning layer, real execution, failure→fix production, no snapshot without contract, no assertion on self-set mock, negative companion) to stop agents from gaming tests via mocks, snapshots, or happy-path-only coverage.
- LLM/agent eval alignment: Connects conventional test rigor to probabilistic LLM oracles with an oracle ladder and eval-dataset principles.
Quick Start
Ask an AI assistant to review or generate your next test for a specific bug by requiring it to state the INVARIANT and OWNING_LAYER first, then check for a negative companion and real execution before providing any test code.