wm-selfmodel — predictive introspection for WhiteMagic.
Tracks per-subsystem metrics over time, forecasts threshold crossings, and feeds confidence signals back into the dispatch pipeline.
Architecture:
- [
MetricTracker] — per-metric ring buffer history with EWMA - [
ForecastEngine] — linear extrapolation + EWMA forecasting - [
AlertEngine] — threshold rules checked against forecasts - [
ConfidenceCalibrator] — overall system confidence from forecast accuracy - [
SelfModel] — top-level orchestrator
The self-model is read-only from the dispatch pipeline (no feedback loops). Confidence <0.5 triggers conservative dispatch (prefer cached results).