Ictal EEG as a Redistribution of Local Synchronization Extremes

The WCM-T3 two-layer reader (companion papers, DOI 10.5281/zenodo.22122562 and10.5281/zenodo.22123404) represents a real-valued feature series as occupancy of the balanced-ternary4-cube: an activity layer counting length-4 windows in which three or more adjacent extrema co-occur (thek≥3 sector), and a sign layer reading the same-sign concordance of exactly-two-extremum windows (the k=2sector). We apply it to scalp EEG from CHB-MIT (six patients, thirty seizures) through a localphase-locking-value (PLV) feature, with a healthy-EEG positive control (PhysioNet EEGMMIDB, n=109).Three results form one chain. (Validation.) On resting healthy EEG the activity layer reproduces the Bergereffect — eyes-closed alpha enhancement, occipital-dominant — with a per-electrode spatial correlation ofr=0.94 against alpha-power physiology. (Mechanism, the central result.) At matched mean synchrony, ictalwindows still show higher activity (weighted gap +0.15; within-PLV permutation p=0.0002; 9/9 PLV bins), sothe reader does not track the level of synchrony but the short-timescale temporal concentration of its extrema;a distribution- and autocorrelation-preserving surrogate is not exceeded, locating the effect at second-ordertemporal persistence (ictal lag-1 autocorrelation 0.248 vs 0.169). The activity gain coincides with depletion ofthe sign substrate (k≥3 vs k=2 occupancy anti-correlated at r=−0.96), so one seizure-associatedreorganization drives a robust activity increase and a weaker, substrate-limited sign change. (Replication.)The direction generalizes across all six independent patients (patient-level sign test and cluster sign-flippermutation p=0.031; one-sample t on patient means t(5)=7.6, p<0.001), stable to leaving any one patient out.We frame this as a mechanistic account, not a seizure detector: that seizures involve synchronization is wellknown; the contribution is that WCM-T3 reads the configuration of synchronization rather than its magnitude,and that the reader itself has a state-dependent observability structure. # Changelog ## v1.1.0 — 2026-08-29Manuscript wording precision and one added robustness analysis (no change tofigures, mechanism, or central conclusions). Prompted by external adversarial review. - **Central claim precision.** "configuration, not magnitude" → "temporal configuration beyond mean magnitude" throughout (the IAAFT surrogate preserves the marginal distribution, so the activity layer is not claimed to ignore magnitude entirely; it captures temporal configuration beyond the mean level).- **Permutation unit (new analysis).** The within-PLV permutation shuffles the ictal label at the window level; since overlapping sliding windows are not independent, added a conservative seizure-block permutation (circular shift of each seizure's ictal mask within its own record). Gap and significance unchanged (both p=0.0002). New script: `epilepsy/permutation_unit_check.py`.- **Frequency-band wording.** "a neural origin is thus favored" → "a simple high-frequency myogenic explanation is thus disfavored" (consistent with the ICA-not-performed limitation).- **Construct validity.** Figure 1 caption "establishing device validity" → "supporting construct validity of the reader" (avoids a clinical-device reading).- **Cohort limitation.** Added a sentence: the cohort was intentionally enriched for clear, well-generalized seizures; patient-level consistency should not yet be extrapolated to subtle or focal phenotypes. ## v1.0.0 — 2026-08-28Initial release (DOI 10.5281/zenodo.22146711). ※ The license for codes is AGPL3.0-or-later