Explore a first-observation scenario
Change the known-at-prediction inputs to explore model sensitivity. Inputs stay in your browser. This does not retrain the model or show causal effects.
Probability of remaining unresolved
A population survival estimate for this feature profile. No calibrated individual confidence interval is claimed.
Out-of-time test results
| Candidate | IPCW Brier ↓ | Event MAE |
|---|
Event MAE uses only observed resolutions within seven days. It excludes censored and longer cases; medians beyond seven days are capped. The hazard model improves the probability score but may not improve time error.
Critical versus moderate incidents
Error by priority
| Priority | Test cases | Observed events | Early censored | IPCW Brier ↓ | Event MAE |
|---|
Small priority groups have unstable estimates. IPCW weights are estimated from training censoring times and assume censoring is independent of resolution behavior; this assumption may not hold.
Training, censoring, and limitations
The model's first-resolution target differs from the main dashboard's latest closing-cycle duration. The original incident dashboard remains unchanged in its metric definitions.