Polyphonic Stationarity: Detecting Regime Shifts in Financial Time Series via Frequency Decomposition, Synthetic Drift Generation, and Lucid Judgment
Author: Henry Carstens Date: April 10, 2026 Pages: 8 Project: Stationarity Project Full paper: PDF
Abstract
Introduces the Polyphonic Stationarity Metric. Decomposes time series via FFT into four voices at target frequencies (252, 126, 63, 31 trading days — a harmonic series). Computes a Polyphonic Drift Score (PDS, 0–1) based on amplitude stability, phase coherence, inter-voice coupling, and consonance ratios interpreted as musical intervals.
Central conclusion: stationarity breaks are not forecastable from market data — either internal dynamics or external cross-market “spiderweb” signals — but are reliably detectable after the fact.
Three falsifications
| Test | Hypothesis | Result | Key finding |
|---|---|---|---|
| Orbital | Internal topology predicts drops | Falsified | Bounded noise |
| Tuning | Autocorrelation predicts drops | Falsified | Independent but non-predictive |
| Spiderweb | Cross-market web predicts target drops | Falsified (0% lift) | Consonance is local |
SFF heuristics confirmed
- SFF1 (Frequency-Specific Stationarity): Stationarity is per-voice, not aggregate. Global tests (ADF) mask this.
- SFF2 (The Keynote): Longest-horizon pair maintains 100% consonance (unison/octave).
- SFF3 (Mid-Range Fragility): Breaks originate in middle frequencies (63–126), where tritones and beating occur.
Spiderweb test results
5-asset FIIJ panel (copper, natgas, TNX, EURUSD, QQQ) vs. USD target:
| Metric | Value | Status |
|---|---|---|
| Best lead correlation | 0.062 | FAIL (1st percentile vs null 95th at 0.171) |
| Zero-lag correlation | 0.015 | Near zero |
| Predictive lift | -0.005 | FAIL |
| Per-asset correlations | -0.029 to +0.044 | Noise |
Consonance is a strictly local property of each market’s internal frequency spectrum. No structural propagation across assets at measurable lags.
Real series baselines
| Dataset | PDS | Amp Stability | Phase Coherence | Coupling |
|---|---|---|---|---|
| USD Daily | 0.42 | 0.64 | 0.60 | 0.03 |
| SPY Hourly | 0.33 | 0.38 | 0.60 | 0.00 |
| CPI Monthly | 0.55 | 0.61 | 0.93 | 0.10 |
All below 0.75 threshold, consistent with regime shifts.
Lucid Judgment application
LJ1–LJ7 governed the entire research program. LJ1 raised evidence bar to require beating FIIJ cascade (41.2% hit rate). LJ2 revised confidence from 35–45% (pre-test) to 5–10% for prediction, >80% for post-facto detection. LJ3 overweighted disconfirming evidence. LJ4 crossed threshold after spiderweb test — retired prediction-seeking. LJ6 confirmed no epistemic collapse. LJ7 activated return trigger on prediction orthodoxy.
Operational tools
- Polyphonic Drift Score (PDS): composite metric, ≥0.75 = stationary, sharp drop = drift
- Lag Sum Estimator: total lag = statistical + fidelity + cognitive tax
- Conservation Triple Audit: theory/measurement/bias layers per LJ6
- Atomic Fragility Flip: triggers on independence collapse
- Drift-Decay Parallel Plot: lockstep confirmation of regime change
Connections to other work
- Grounded in heuristic-algebra — uses ⊕ha, ⊕inn, ¬, π operators
- lucid-judgment as meta-governance for the research program
- Synthetic data generation follows SDG1–SDG5 axioms
- Musical analogy from Heuristics of Harmony (Ha1–Ha7)
See Also
- polyphonic-stationarity — the concept page
- heuristic-algebra — the formal system
- lucid-judgment — meta-reasoning framework used throughout