oliverhennhoefer/nonconform

Turn anomaly scores into decisions with a stated error rate using nonconform, a Python library for conformal anomaly detection. Use this skill whenever someone wants to flag anomalies, outliers, novelties, fraud, defects, intrusions or out-of-distribution points and the number of false alarms matters, for example when choosing an anomaly threshold, replacing a contamination rate or top-k cutoff, controlling the false discovery rate (FDR) or false positive rate, getting p-values for anomaly scores, calibrating scores from PyOD, scikit-learn, an LLM judge or embedding distances, or detecting when a data stream changed with a bounded false alarm probability. Also use it whenever code imports nonconform.