Conceptversion drift
Version Drift
A common problem in software engineering referring to the gradual invalidation of historical measurements or configuration results as a system undergoes iterative versioning. This is particularly hazardous in AI evaluation contexts, where it can cause hidden discrepancies between published metrics and actual performance.
Timeline (last 90 days)
Sep 17
Version drift is dangerous in evaluation scenarios because the validity of historical measurement results implicitly decays after changes to model weights, inference frameworks, or test environments
Unverified50%