Online/Offline Skew
A critical failure mode where a model's performance degrades in production because the real-time inference data differs from the batch training data.
Think of It Like This
Like practicing for a spelling bee using a written dictionary, but failing on stage because you can't understand the announcer's accent.
This skew often arises from subtle bugs in feature engineering pipelines, where a feature is calculated slightly differently in the real-time streaming engine compared to the historical data warehouse. It can also occur due to delayed data availability in production. Robust MLOps practices, like unified feature stores, are required to prevent this.