SCPN Phase Orchestrator: Fact-Based Overview

This page states, plainly, what SPO is, what has been checked against an independent ground truth, and what has not. It deliberately avoids superlatives and does not describe review-only or unvalidated surfaces as if they were deployed capabilities.

1. What SPO is

SPO is a Python library and CLI for analysing coupled-oscillator systems. It turns the repeating signals a system produces into phase variables built on Kuramoto / UPDE dynamics, and provides detectors, a regime supervisor, and a review-only control-proposal surface. It is a research and evaluation toolkit, not a deployed controller and not a validated general early-warning system.

At its core is the Universal Phase Dynamics Equation (UPDE):

\[\dot{\theta}_i = \omega_i + \sum_j K_{ij}\,\sin(\theta_j - \theta_i - \alpha_{ij}) + \zeta\,\sin(\Psi - \theta_i)\]

2. Evidence status (validated vs not)

Externally validated against an independent reference:

• Grid modal damping estimation. On the IEEE-39 and Kundur systems, the estimated growth rate of the dominant electromechanical mode matches the small-signal eigenvalue from the ANDES simulator (study §3.9).
• The eigenvalue regime map. Across five systems (fold, pitchfork, Hopf, and the unimodal and bimodal Kuramoto transitions) the shipped detectors recover the analytic eigenvalue's real part; the correct estimator is regime-dependent (study §3.10–3.14).
• Honest, false-alarm-controlled evaluation. A matched-false-alarm operating point, a permutation significance test, and a hash-sealed evidence record (study §2).

Empirically at chance on real data: across five real modalities (grid, EEG, ecological/climate, molecular), generic early-warning detection at an honest operating point is at chance under a permutation-controlled test (study §3.1–3.8), consistent with the wider literature. SPO does not claim to predict tipping points in these domains. The one place a detector clears the bar is the grid.

Not yet validated: closed-loop control, hardware / PLC / FPGA / quantum / neuromorphic actuation, BCI feedback, and distributed-mesh coupling are adapter-scoped, review-only, and carry no field evidence.

3. Components that exist in the codebase

These modules are present and covered by tests. Presence is not a performance or deployment claim; several are optional or review-only.

Area	Module	What it does
Dynamics	upde/	UPDE / Kuramoto integration, bifurcation and Ott–Antonsen reduction
Differentiable	nn/ (JAX/equinox)	phase-oscillator layers, order parameter, coupling inference; can run on GPU/TPU
Detection	monitor/	critical-slowing-down indicators and grid modal growth-rate estimation
Evaluation	evaluation/	detector-agnostic matched-false-alarm + label-permutation auditor with hash-sealed records (spo audit-detector)
Supervision	supervisor/	regime classification (nominal / degraded / critical) and a Petri-net state machine
Control (review-only)	policy surfaces	bounded, rate-limited control proposals with replay evidence — not actuation
Assurance	assurance/	hash-sealed evidence records and canonical-record hashing
Acceleration	spo-kernel (Rust FFI)	integration kernels with parity tests against the Python path
Adapters (optional, review-only)	adapters/	OPC-UA / MQTT ingestion, LSL, mesh, and simulator bridges — scaffolds, not validated integrations

4. Performance claims

Latency and throughput are environment-specific and are not quoted here as static facts. Where a number is needed, reproduce it on the target hardware with the committed benchmark scripts and record the run metadata (platform, Python version, backend, lockfile). The canonical benchmark snapshots under benchmarks/ and the CI benchmark gates are the reference; treat any figure without attached run metadata as unverified.

5. Where SPO fits

SPO is between an academic research toolkit and a production monitoring system. Its one externally-validated niche is grid modal-damping estimation checked against eigenvalues; its most transferable asset is the honest, false-alarm-controlled evaluation methodology, which applies to any early-warning claim. It is a complement to ground-truth tools such as ANDES (which it validates against), not a replacement for deployed control-room monitoring products.

For related material, see the study and the architecture and product-boundary pages.