Auditing Individual Specificity in EEG Foundation Models with Occupancy-Preserving Temporal Nulls

EEG foundation models are typically evaluated through downstream predictive performance, providing limited insight into what individual-specific information their frozen representations preserve when transferred to an unseen dataset. In this study, we introduce a representation-audit framework that distinguishes fingerprintability from temporal fingerprintability. A temporal representation supports an individual-specific dynamical interpretation only if its cross-session fingerprintability exceeds an occupancy-preserving temporal null. We formalize this distinction through temporal excess, ${\Delta }_{\mathrm{temp}}(g)={\Delta }_{\mathrm{FP}}^{\mathrm{real}}(g)-\mathbb{E}[{\Delta }_{\mathrm{FP}}^{\mathrm{shuffle}}(g)]$. On the Temple University Epilepsy Corpus (TUEP), we evaluate two independently pretrained frozen EEG encoders, CNN–Transformer and LUNA-Base, and discretize their latent representations into $k=4$ states. State occupancy is reproducible across separate recording sessions ($N=112$), with ${\Delta }_{\mathrm{FP}}=0.103$ for CNN–Transformer and $0.096$ for LUNA-Base. Empirical lag-1 transition representations $W$ also appear individually specific under conventional metrics, reaching Top-1 retrieval rates of $5.8\%$ and $4.5\%$, respectively, compared with $0.9\%$ chance. In contrast, the apparent temporal fingerprintability does not exceed the occupancy-preserving temporal null: ${\Delta }_{\mathrm{temp}}(W)=-0.002$ for CNN–Transformer (real vs. shuffled ${\Delta }_{\mathrm{FP}}$: $0.106$ vs. $0.108$) and $-0.010$ for LUNA-Base ($0.095$ vs. $0.105$). Residual-transition and state-persistence probes yield the same conclusion at the primary $k=4$ resolution. These results show that a temporal representation can be individually fingerprintable without its fingerprintability depending on temporal order. More generally, ${\Delta }_{\mathrm{FP}}>0$ does not imply ${\Delta }_{\mathrm{temp}}>0$, motivating occupancy-preserving temporal nulls when interpreting individual-specific dynamics in learned representations.