Mind · Inference
RegimeSense
Detecting structural breaks in reality
Definition
Mechanism
An anomaly detector that continuously tracks divergence between the agent's internal model predictions and observed reality. It calculates a Regime Stability Score (RSS) as the rolling average of prediction accuracy over a window. High surprisal (low accuracy) indicates regime instability. When RSS drops below threshold, it signals that the underlying generative process has changed. It continuously monitors prediction error via DriftWatch, triggering OntologyAdapt or Quorum if the error signal crosses the regime-change threshold.
Preconditions
- Agent has a predictive model generating expectations
- Observations are timestamped and comparable to predictions
Postconditions
- Alert raised if regime break detected
- RSS score computed for current window
Invariants
- Stability Break: RSS below `stability_threshold` across the window signals a regime break.
- The RSS window is `window_size` observations, fixed in advance rather than chosen per evaluation.
Risks
Failure modes
False Positives
Noise spikes triggering unnecessary accommodations.
Lag
RSS is a trailing indicator, so regime shifts are detected after-the-fact.
Threshold Gaming
Setting threshold too high causes missed shifts; too low causes constant alerts.
Relationships
Builds on
References
Autonomous Goal-Directed Entity
Infrastructure
Failure-announcement signal {severity, reason, recoverable}, ACK-required
Society
Reputation scoring via behavioral deviation from baseline (2-sigma alert)
Physics
Irrelevant information
Restructuring categories to fit data
Infrastructure
Minimum participation threshold
Used by
References
Know where your framework stops working while lacking a replacement, and use it only inside that edge
Learning weighted by prediction failure magnitude
Reference
Parameters
stability_thresholdFloat, [0.0, 1.0]- RSS threshold below which regime break is detected
window_sizeInteger, [10, 1000]- Number of observations in the rolling average window