The Fracttalix Meta-Kaizen Series

# Fracttalix Sentinel v7.3 **Lightweight, regime-aware anomaly detection for time series** Single-file Python | No dependencies required | CC0 public domain Sentinel is a simple, fast anomaly detector for univariate or multivariate time series. It uses adaptive EWMA thresholding + bidirectional CUSUM to spot regime shifts early, making it especially useful for: - Finance (volatility regime changes, BTC/asset monitoring) - Physiology (HRV, vital sign monitoring) - IoT & infrastructure (sensor drift, network anomalies) - Research & exploratory analysis (quick screening with surrogates) ### Current Version: v7.3 (March 2026)- Two-sided alerts (upper & lower anomalies)- Per-channel multivariate detection with flexible aggregation- Soft/full regime reset (preserve baselines during gradual drifts)- NaN/Inf imputation (skip or mean)- State persistence (JSON save/load for long-running sessions)- Volatility-adaptive smoothing- Optional Numba JIT acceleration for fractal/entropy metrics (if installed)- Optional parallel surrogate generation & tqdm progress- Proper FFT-based phase randomization for surrogates- Buffered CSV export (low filesystem overhead: accumulate in memory, flush on interval or manually) ### Quick Start```pythonfrom fracttalix_sentinel import Detector_7_3 detector = Detector_7_3( alpha=0.12, early_mult=2.75, fixed_mult=3.2, two_sided=True, per_channel_detection=True, volatility_adaptive=True, verbose_explain=True, csv_output_path="alerts.csv", # Optional: save alerts to CSV csv_flush_interval=60 # Flush buffer every 60 alerts (0 = manual only)) # Feed data (scalar for univariate, list for multivariate)for value in your_time_series: result = detector.update_and_check(value) if result.get("early_alert"): print("Early alert:", result) if result.get("confirmed_alert"): print("Confirmed alert:", result) # Manual flush (optional)detector.flush_csv()