The generic GP is a tutorial active-learning surrogate for one-dimensional saturation-dose experiments. It is used to compare a stationary data-driven model against fixed grids and physics-informed residual models.
- Precursor dose in seconds.
- Observed growth response from Stage 1 virtual process families.
- Configured noise mode and candidate grid.
- Posterior mean and uncertainty on the candidate dose grid.
- Threshold-oriented candidate proposals.
- Tested-dose recommendation when available.
- Smooth response over a bounded one-dimensional dose interval.
- Stationary kernel structure is adequate enough for tutorial comparison.
- Observations are generated by the educational simulator, not a real reactor.
- It does not encode saturation physics directly.
- It can be inefficient on slow or soft saturation curves.
- Calibration can degrade under high noise or model misspecification.
Runs record seeds, settings, warnings, observations, and recommendations. The model is deterministic for fixed seeds and dependency versions.