User Guide

In-depth coverage of algorithms, cost models, the doctor engine, and practical workflows.

• Use Cases
  • Industrial sensor monitoring (mean-shift detection)
  • Financial regime detection (variance change)
  • Network traffic anomaly detection (streaming BOCPD)
  • A/B testing (detecting treatment effect onset)
  • Climate data (seasonal + trend)
• Offline Algorithms
  • PELT (Pruned Exact Linear Time)
  • BinSeg (Binary Segmentation)
  • FPOP (Functional Pruning Optimal Partitioning)
  • SegNeigh / DynP (Exact Dynamic Programming)
  • WBS (Wild Binary Segmentation)
  • BottomUp
  • SlidingWindow
  • KernelCpd (experimental)
  • GpCpd / ArgpCpd (experimental)
  • Stopping criteria
• Online Algorithms
  • BOCPD (Bayesian Online Changepoint Detection)
  • CUSUM (Cumulative Sum)
  • Page-Hinkley
  • Common patterns
• Cost Models
  • Overview
  • Per-model details
  • Decision tree: choosing a cost model
  • Availability in Python
• Doctor Recommendation Engine
  • What the doctor does
  • CLI workflow
  • Python integration
  • Objectives
  • Calibration families
  • Confidence formula
  • Preprocessing recommendations
  • Worked example
  • Multivariate awareness
• Preprocessing
  • Pipeline stages
  • Full configuration example
  • Stage details
  • Validation rules
• Reproducibility
• Reproducibility Modes and Deterministic Contracts
  • Scope
  • Mode Contract Matrix
  • Strict Mode
  • Balanced Mode (Default)
  • Fast Mode
  • Per-Cost-Model Score Tolerances
  • Cross-Platform Reproducibility
  • Configuration Examples
  • FAQ
• Serialization
  • Result JSON contract
• Offline Result JSON Contract
  • Canonical Schema + Version Marker
  • Top-Level Field Contract
  • Diagnostics Contract
  • Backward/Forward Compatibility Expectations
  • Validation Failure + Error Messaging Contract
  • Compatibility Fixtures
  • Online detector checkpoints