copy-paste

This record (Part II) presents SS3 (Ss3), a deterministic structural operator for event/episode detection in noisy 1D signals. SS3 maps an input signal F to (G, M, E) using |ΔF|, EWMA memory, and quantile thresholds (optional hysteresis) to output invariant events/episodes.

Core (validated parameters):
G[i] = |F[i] − F[i−1]|,  G[0]=0
M[i] = α M[i−1] + (1−α) G[i],  M[0]=0
θ = Q_q(M),  E[i] = 1[M[i] > θ]
Hysteresis: θ_h = Q_{q_h}(M), θ_l = Q_{q_l}(M), q_l < q_h; enter M>θ_h; stay M>θ_l; exit otherwise.
Fixed validated params: α=0.95, q=0.95, q_h=0.95, q_l=0.93.

JWST/NIRISS (NIS_SOSS, Level 2b *_x1dints.fits) results reproduce 102 invariant events across segments (seg001–seg003) under fixed parameters; under 5% additive noise, event count remains 102 with agreement ≈ 0.9205103042. A comparison with find_peaks shows peak counts increase under noise while SS3 remains structurally stable. TDA (delay embedding) summary is included as a structural consistency check. A transit-boundary demo (ingress/egress) is provided on a wavelength-bin lightcurve; this is a boundary-detection demonstration and NOT an official new-planet discovery claim without independent confirmation.

BTC forward test (BTCUSDT 1m, N=4999, shocks=36, warning_enters=11) evaluates warning_enter precision for shocks defined by rolling quantile on |Δlog(close)| with q_shock=0.995. Results (T15): H=5 precision_enter 0.273 vs baseline_mean 0.035 (lift 7.762, p=0.005); H=10 0.455 vs 0.064 (lift 7.148, p≈0.000); H=20 0.545 vs 0.109 (lift 5.021, p=0.001). Interpretation: statistically significant early-warning for imminent volatility shocks (not price direction).

Related work (Part I): DOI 10.5281/zenodo.18779582Code (GitHub/Zenodo archive): DOI 10.5281/zenodo.18821006
Repository: https://github.com/ogamilife3-a11y/S3-Algorithm