Production Experiment Cost Attribution with Feature Flags and Incident Rollbacktranslating…

Short answer: put the kill switch in a small control plane, have every Node.js worker evaluate a locally cached flag before entering the expensive experiment path, and attribute cost by tenant cohort, flag revision, and evaluation version so an incident rollback stops new work…