This is a ready-to-use log for the ongoing performance monitoring of a digital twin. A twin validated at release can degrade silently as raw materials, equipment, and conditions drift, with no change to the twin at all. The validated state therefore includes monitoring, and this log is the evidence that the twin still works. It records the twin’s prediction against the actual measured outcome, tracks error and bias over time, and captures the response when a threshold is tripped. Fill one row per batch or per defined interval. Replace every <<FILL: ...>> placeholder with your own specifics. A filled specimen follows. Verify each cited standard against the current source before you rely on it.
Log control
| Field | Entry |
|---|---|
| Log title | Digital Twin Performance Monitoring |
| Document number | <<FILL: LOG-ID, e.g. LOG-TWIN-005>> |
| Version | <<FILL: version>> |
| Twin name / ID | <<FILL: TWIN NAME / ID>> |
| Predicted variable and units | <<FILL: e.g. product titer, g/L>> |
| Reference method (record of truth) | <<FILL: e.g. offline titer assay>> |
| Owner | <<FILL: System Owner / Process Science>> |
Monitoring parameters
| Parameter | Value | Basis |
|---|---|---|
| Monitoring frequency | <<FILL: per batch / per interval>> | <<FILL>> |
| Error metric | <<FILL: absolute error, RMSE window, or MAE>> | Matches the validation acceptance metric |
| Error threshold (action limit) | <<FILL: value>> | Tied to what the decision tolerates |
| Bias threshold | <<FILL: value, directional>> | A systematic drift is a specific hazard |
| Trend window | <<FILL: e.g. rolling 10 batches>> | To catch slow drift, not just single points |
Field definitions
| Field | Format | Required | Who | When |
|---|---|---|---|---|
| Batch / interval ID | text | Yes | System Owner | Each batch |
| Date | date | Yes | System Owner | Each batch |
| Model version | text | Yes | System Owner | Each batch |
| Twin prediction | number | Yes | System | At the decision point |
| Actual (reference method) | number | Yes | Lab | When the reference result returns |
| Error (actual minus prediction) | number | Yes | System Owner | On reference result |
| Within threshold? | Yes/No | Yes | System Owner | On reference result |
| Trend flag | text | Yes | System Owner | Each batch |
| Action taken | text | Yes | System Owner | If threshold tripped |
| Reviewed by / date | text | Yes | Process Science / QA | Per review cadence |
Log
| Batch/Interval | Date | Model ver | Prediction | Actual | Error | Within threshold? | Trend flag | Action | Reviewed by/date |
|---|---|---|---|---|---|---|---|---|---|
<<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
<<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
Defined response when a threshold trips
When a single error exceeds the action limit, or the trend window shows a sustained bias or rising error, the defined response is:
<<FILL: e.g. suspend the twin's advisory role and revert the decision to the reference method>>.- Raise a deviation or investigation per
<<FILL: SOP-ID>>and determine root cause (raw material lot, cell line, equipment, sensor calibration, model drift). - Do not return the twin to its GxP role until the cause is understood and, if the model changed, revalidation under the change plan is complete.
A monitoring log with no defined response is decoration; the response is what makes the monitoring a control.
Retention and review
Retain the log for the life of the twin plus the applicable records-retention period. Review it at the defined cadence and feed a persistent drift into continued process verification and the periodic review of the twin.
References
ICH Q9(R1), Quality Risk Management; ICH Q10. 21 CFR Part 11; EU GMP Annex 11 (Computerised Systems, in-force 2011 version).
Confirm the current version of each reference before issue.
Filled specimen
The following shows three rows for an example titer soft-sensor twin, so you can see the expected detail. The values are illustrative; the error threshold here is a stated absolute error and the bias threshold a sustained one-directional drift.
| Batch/Interval | Date | Model ver | Prediction | Actual | Error | Within threshold? | Trend flag | Action | Reviewed by/date |
|---|---|---|---|---|---|---|---|---|---|
| BR204-2607-018 | 03-Jul-2026 | 2.1 | 4.10 g/L | 4.18 g/L | +0.08 | Yes | normal | none | M. Ito, 04-Jul-2026 |
| BR204-2607-019 | 05-Jul-2026 | 2.1 | 3.95 g/L | 4.05 g/L | +0.10 | Yes | watch: 2nd consecutive under-read | none | M. Ito, 06-Jul-2026 |
| BR204-2607-020 | 07-Jul-2026 | 2.1 | 3.80 g/L | 4.02 g/L | +0.22 | No (limit 0.15) | trend: 3rd under-read, bias rising | Advisory role suspended; reverted to offline assay; DEV-2026-0210 raised for new raw-material lot | M. Ito, 08-Jul-2026 |
In this example the log caught a drift before it caused a bad decision: two in-limit but same-direction errors were flagged as a watch, the third breached the action limit, and the defined response fired, the twin’s advisory role was suspended, harvest reverted to the offline assay, and a deviation was opened that traced the drift to a new raw-material lot. That sequence, watch to breach to defined response, is what turns monitoring into a control an inspector accepts.
Common inspection findings this log prevents
- Deploy and forget: no monitoring, so there was no evidence the twin still predicted accurately.
- Monitoring that recorded numbers but had no threshold, so nothing ever triggered a response.
- A sustained one-directional bias that no one flagged because only single-point errors were checked.
- A degrading twin that kept informing decisions because there was no defined response when error tripped.
How to adapt this log
- Set the error and bias thresholds to match the validation acceptance criteria, not looser values.
- Track a trend window, not just single points, so slow drift is caught before it breaches.
- Write the defined response before you need it, and make suspension of the twin’s role the default first step.
- Keep the reference method running so you always have a record of truth to compare against.
- Confirm every standard in the references against the current published version before issue.