Why your conformance rate always reads 100%
A conformance check that replays a log against a model discovered from that same log cannot find anything. It is a classic process-mining trap, and an easy one to spot.
Conformance checking compares what happened with what should have happened. That only works if the two come from different sources. When the tool discovers the model from the event log, then replays that same log against it, it is checking that the data looks like the data. The answer is known in advance.
What a discovered model actually measures
A discovery algorithm produces a model that accepts the observed behaviour: that is its job. Replay can then only flag cases excluded by the simplification thresholds: statistical noise, not business deviations.
What is needed instead
A normative model: the procedure as written, imported from a BPMN file, a quality standard or an internal reference. It is the only baseline against which a deviation means something.
- The model comes from somewhere other than the log.
- It describes intent, not observation.
- Its deviations read in business terms: a skipped step, a reversed order, a rework loop.
So the fix is not a threshold adjustment, it is a change of source. As long as the baseline comes out of the log, no setting will make the result usable.
