methodSystems and networks

Change-point detection

Detect candidate shifts in level, variance, trend, or model parameters with controlled false alarms.

Working definition

Change-point methods search an ordered series for times where a statistical property changes. Offline methods segment a completed record; online methods monitor sequentially. Penalties, priors, minimum segment lengths, and detection delays determine what counts as a change.

Notation

τ = argmin segmented loss + penaltyH₀: θ₁ = θ₂; H₁: θ₁ ≠ θ₂

Assumptions

  • The changing property is defined.
  • Noise and dependence are modeled.
  • Detection thresholds are selected without outcome cherry-picking.

Invariants

  • A detected point is conditional on model and threshold.
  • Online detection occurs after evidence accumulates.
  • Multiple candidate points consume an error budget.

Reproducible procedure

  • Specify change type and minimum duration.
  • Fit null and segmented alternatives.
  • Validate with sensitivity, false-alarm simulation, and external evidence.

Error and boundary controls

  • Gradual drift can resemble multiple abrupt changes.
  • Outliers can trigger false points.
  • Small samples make location uncertainty wide.

What this does not establish

A change near a transit or ingress does not establish correspondence beyond chance; event windows and controls must be pre-registered.

Explicit applications

2 cross-domain bridges

Semiconductor manufacturingmeasurement

Tool and lot regime shifts

Detect candidate shifts in defect rate, critical dimension, or equipment telemetry.

Inputs

  • ordered fab measurements
  • lot boundaries
  • maintenance events

Outputs

  • candidate shift times
  • location uncertainty
  • external-evidence checklist

Transformation: Compare stable and segmented process models under a preselected false-alarm policy.

Limit: A detected shift localizes a change but does not identify the responsible chamber, material, or recipe.

Open connected system →
Empirical validationempirical test

Outcome regime-change audit

Detect business or measurement shifts independently before testing whether registered celestial windows add information.

Inputs

  • objective outcome series
  • operational covariates
  • detection protocol

Outputs

  • change windows
  • location uncertainty
  • non-celestial explanations

Transformation: Identify candidate changes under a controlled false-alarm process.

Limit: Temporal coincidence with a planetary event is not sufficient evidence of correspondence.

Open connected system →

Authoritative references

  1. [1]NIST/SEMATECH e-Handbook of Statistical Methods · National Institute of Standards and Technology

    Establishes: Methods for uncertainty analysis, calibration, time-series modeling, process monitoring, experimental design, reliability, and statistical comparison.

    Boundary: Statistical procedures quantify evidence under a design and model; they do not repair biased sampling, outcome leakage, post-hoc hypotheses, or unmeasured confounding.

Related mathematical concepts