non-equilibrium-auditor

Time-Series Event Auditing for Structural Collapse Detection

non-equilibrium-auditor is an experimental toolkit designed to detect and audit non-equilibrium transitions and structural collapse events in complex, noisy multi-channel time-series data.

Unlike traditional anomaly detection, this system focuses on the regime break—the moment a system crosses a stability barrier and enters a collapsed state.

---

Concept: Structural Collapse

Many real-world systems remain in a stable basin until a perturbation pushes them across a barrier. This transition is rarely a simple spike; it is a structural shift characterized by:

• Precursor Activity: Subtle signs of instability before the main break.
• Winner Event: The primary transition that marks the regime shift.
• Non-Equilibrium Asymmetry: A physical "arrow of time" that makes the collapse irreversible.

---

System Architecture

The pipeline separates signal analysis into a Dual Branch architecture to isolate the event from its environment.

LF Branch — Environment Map

Tracks slow environmental behavior, baseline drift, and long-term oscillations. It acts as a reference map to ensure environmental noise isn't mislabeled as a structural event.

HF Branch — Transition Scan

Searches for rapid structural changes. This branch evaluates potential candidates using metrics like local SNR, irreversibility, and persistence.

---

Audit & Classification

Detected candidates are categorized into a discussable and inspectable hierarchy:

• Winner: The strongest candidate transition near the target window.
• Precursor: Supporting transitions that precede the main event.
• Artifact / Environment / Noise: Rejected candidates based on symmetric artifacts, narrowband contamination, or long-term drift.

---

Event Metrics

Each event is audited using interpretable physical metrics:

• P_snr: Local signal contrast against surrounding noise.
• T_irreversibility: Measures the transition-like asymmetry between rising and falling phases.
• Persistence: Counts supporting adjacent windows (hit_windows) to ensure the event has physical substance.
• Artifact Ratios: Rejects symmetric patterns (Ratio_MirrorSym) often produced by sensor glitches or DSP side-effects.

---

Components

1. Python Engine (death_audit_v15.py)

The core backend that processes raw multi-channel data and ranks events.

2. Interactive UI (index.html)

A browser-based analysis dashboard to visualize, verify, and audit detected events without additional software.

---

License

© DIMProductions Research use only.