Predictions and outcomes
Separate a future estimate from confidence, execution and later observed results
A forecaster defines a repeatable question about the future. A prediction records one answer for a subject and a window. An evaluation records what actually happened. The three records let you inspect an answer and keep a track record without rewriting the original prediction.
What a forecaster contains
It combines a published ontology evidence-query release, instructions, input/output schemas, a model, execution configuration and evaluation policy. It belongs to a project. Each executable version pins its evidence release; publishing another query release does not silently move that pin.
Four kinds of answer
| Kind | Example | Supported score |
|---|---|---|
| Probability | Chance of a defined equipment failure within seven days | Brier score and aggregate calibration |
| Category | Maintenance priority at an evaluation date | Exact-match accuracy |
| Number | Unplanned downtime hours during a window | Absolute and squared error |
| Scenario | Description of possible failure conditions | Recorded outcome/notes; no built-in numeric scenario score |
Probability is the estimated likelihood of the event. Confidence is a separate model-generated assessment of support. Neither is an observation that the event happened.
Example: equipment failure
Define failure as an unplanned mechanical failure causing at least one hour downtime within seven days. A run for P-101 loads its pinned query evidence and saves an output. Completed means execution/output validation succeeded; its outcome can still be Pending.
A real occurrence can settle a probability event before the window ends. Non-occurrence requires the closed window and adequate monitoring. Missing evidence is Unknown, not Did not occur. Evaluations keep observation time and any assertion references as versioned records.
Recording outcomes does not automatically retrain the model. Measure compatible forecasts against real observations and inspect missing outcomes before interpreting an aggregate report.
Build and evaluate a prediction
Forecasting and predictions covers the full lifecycle. Your first equipment prediction includes source, ontology, evidence query and forecaster setup. Outcome evaluation explains timing and observations. The deployment must have forecasting and its model/workflow runtime configured.