Failure mode
Schema Drift
An unplanned structural change to data that alters downstream behavior, often without immediately breaking the pipeline or the user interface.
Domain: data-movement-reliabilityDomain: change-reliability
ARF layer: context-stabilityARF layer: ai-readiness-interoperability
Signals
- columns appear, disappear, or change type without controlled rollout
- data continues to land, but downstream assumptions are no longer valid
- semantic breakage appears far away from the original source change
Why it matters
Schema drift often presents as a data engineering issue, but it becomes a semantic and decision problem quickly. If column shape changes alter joins, filters, or retrieval context, downstream AI and analytics outputs become less trustworthy even when dashboards keep refreshing.
Prevention patterns
- schema contracts and explicit change review
- automated drift detection before downstream publication
- controlled evolution paths for streaming systems and late-arriving schema changes