Execution Drift
Execution drift is when the same analytical workload behaves differently over time or across environments due to runtime changes rather than intentional logic changes. It often appears as performance regression, different execution plans, or inconsistent latency that changes how users interact with analytics.
Common failure signals
- Query latency changes without code or model changes
- Different execution plans for the same query (or the same query parameters)
- Concurrency limits or resource contention create inconsistent results delivery times
- "Works in dev, times out in prod" for identical logic
Often confused with
- Pipeline freshness issues -- execution drift is runtime behavior, not data timeliness
- Code regression -- drift can occur without any code change
- Observability noise -- drift is consistent enough to affect outcomes and user behavior
Where it shows up in Analytical Reliability
- Execution Reliability: runtime behavior changes across systems, tools, or environments
- Change Reliability: config, scaling, or dependency changes alter execution without touching logic
- Semantic Reliability: users change filters/questions to "make it run," creating intent mismatch