What-if experiments. Each run executes a seeded baseline and the scenario side-by-side, so every delta is attributable to your incidents.