Eval results

Each scenario runs N times against stateful mock apps. A run passes only when an independent oracle finds the expected final app state, not when the agent says it succeeded.

Read the reliability brief (BRIEF.md) ↗(opens in new tab)