A wastewater forecast is useful when it identifies a relevant risk early enough for operators to verify conditions and respond. Assess this on the facility's own records instead of relying on a universal accuracy or warning-time claim.
Define the event and available response
Choose one risk and specify how an event is confirmed. Identify who receives the alert and how much time is useful for inspection. Separate early-warning objectives from the rules used to judge treatment results.
Check data and operational coverage
Review measurement sources, missing periods, sensor checks and operating modes. Link records to confirmed incidents and process changes. A dataset that covers only stable operation may not support evaluation of rare disturbances.
Evaluate on data the model did not train on
Use a review period appropriate to the event and retain the time order of records. Record correct alerts, false alerts, missed events and useful lead time. Check whether results change across operating modes.
Run in observation mode before expansion
Let operators review predictions alongside existing alarms. Log their findings and actions. Agree acceptance criteria and responsibilities before the pilot, and reassess the model when sensors, equipment or operating conditions change.
Records to check
| Record group | Evidence to review |
|---|---|
| Correct alert | Confirmed event and warning time |
| False alert | Inspection result and operational burden |
| Missed event | Confirmed incident without useful warning |
Prepare an assessment with X-IRIS
Bring historical data, verified events, known data gaps and one risk that the operations team wants to investigate. Review the implementation roadmap and request an assessment to agree a scope that can be tested.
