Grounding engine
The subsystem that compares external telemetry against world-model predictions to compute prediction error, correct model confidence estimates, and trigger active inference probes in high-uncertainty areas.
The grounding engine is what keeps the world model tethered to reality. The world model makes predictions about what will happen given a candidate action; the grounding engine reads the actual telemetry after that action and computes how far off the prediction was. This prediction error signal has two uses. First, it directly corrects the world model's confidence estimates: a series of errors on predictions about a specific system or operational class causes the model to lower its confidence in that domain. Second, it identifies high-uncertainty areas where the model is consistently wrong or consistently unable to make confident predictions - these areas are flagged for the curiosity engine as candidates for active investigation. The grounding engine also validates the causal model: if the causal model predicts that intervention A causes outcome B, the grounding engine can check whether historical telemetry supports that directional claim, and can trigger a revision of the causal structure when the data contradicts the model. Without grounding, the world model would drift from reality as the infrastructure it is modeling evolves and its historical training data becomes stale.