Implemented
Strict CSV validation, case grouping, timestamp sorting, directly-follows counts, median elapsed times, exact variant frequencies and event-by-event replay. Invalid IDs, unknown activities and equal times within a case are rejected. No LLM or learned predictive model is connected.
Not a client benchmark
The sample is constructed dispatch telemetry. A frequent path is not automatically correct. Missing events can look like skipped work; event gaps do not prove who caused a delay. Repeated events and incomplete cases need a domain review.
Production next
Authorized ERP/gate/quality event adapters, identity mapping, versioned rules and retention controls. Validate event completeness and clock semantics first. Then test whether grounded explanations or AI suggestions add value. No operational action is triggered.