# Statistics-Based Debugging
**Entity class:** Method / fleet-scale fault analysis
## Definition and history
Statistics-based debugging compares large numbers of error reports to identify recurrent failure patterns and likely causes. [[articles/Intelligence Inside|Intelligence Inside]] uses Microsoft's Windows Error Reporting paper as its example of the method at global software scale. A single event can be ambiguous, while buckets and distributions across billions of reports expose common signatures. The method turns isolated incidents into evidence about a population of installations.
## Mechanism and limits
Its power depends on consistent event definitions, adequate sampling, and enough context to avoid collapsing distinct failure causes into one bucket. A collection system also shapes which incidents become visible to the analyst. The analysis is an institutional inference applied after reports arrive, rather than proof that each machine individually learned the failure model.
## Relationships
**Implemented by:** [[wiki/Windows Error Reporting|Windows Error Reporting]]. **Input:** [[wiki/Telemetry|Telemetry]]. **Related operation:** [[wiki/Classifier|Classifier]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]].
## Sources
[Microsoft Research, *Debugging in the (Very) Large*](https://www.microsoft.com/en-us/research/?p=157688); [[articles/Intelligence Inside|Intelligence Inside]].