REGIME & ANOMALY
Hidden Markov Model 4-state regime detector · Mahalanobis anomaly layer · daily refit
⚠️ Failed to load data — please retry shortly
⏳ Model warming up
The HMM needs ≥12 observations to fit its first parameters and ≥60 observations to be considered fully calibrated.
Current sample: — observations.
Note: archive entries with score = 0 (the Mar 9 → Apr 24 producer-bug period) are filtered from training. As more daily data accumulates from the post-fix Apr 25 baseline, the model retrains and improves automatically.
HMM REGIME PROBABILITIES
probabilistic state membership · current observation
TRANSITION PROBABILITIES
P(next state | current state) · diagonal = stay-in-state probability
FIT QUALITY
model diagnostics · how confident is the HMM in its parameters
SIGNAL ANOMALIES
per-signal Mahalanobis z-score vs 90d rolling distribution