Deliberately testing an algo against rare or extreme scenarios — a flash crash, a stale price feed, a venue outage — rather than only “normal” market conditions. Most execution failures happen in edge cases, which is exactly why they’re worth simulating instead of discovering live.