EEG foundation models achieve state-of-the-art clinical performance yet their internal computations remain opaque, constituting a barrier to clinical trust.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Clinical concepts in EEG foundation models fall into three operational regimes: selectively steerable, encoded but entangled, and non-encoded.claim0.864
  Interpretive claim summarizing the spectrum of concept steerability discovered.
• We hypothesize that applying SAE-based mechanistic interpretability to EEG foundation models can expose representational failures and thereby improve clinical trust.hypothesis0.836
  Overarching motivating hypothesis of the paper
• EEG foundation modelsconcept0.830
  Large transformer models pretrained on EEG data for clinical tasks; the object of interpretability in this paper.
• How are clinical concepts represented and steerable in EEG foundation models?question0.825
  Core research question driving the mechanistic investigation.
• Age and pathology are clinically entangled in EEG foundation model representations such that suppressing one concept inevitably corrupts the other.claim0.820
  A specific representational failure with direct clinical safety implications
• What clinical concepts are encoded in the internal representations of EEG foundation models?question0.820
  Primary research question driving the extraction and benchmarking of SAE features
• Can concept steering interventions on EEG foundation models be made selective rather than globally destructive?question0.785
  Research question motivating the introduction of the probe area metric and identification of operational regimes
• What physiologically interpretable frequency signatures correspond to latent concept steering manipulations in EEG foundation models?question0.769
  Research question motivating the spectral decoder methodology