{
  "abstract": "Pretrained EEG foundation models are increasingly proposed as general-purpose encoders for brain-computer interfaces, yet recent benchmarks disagree about when their representations transfer to downstream tasks. We audit LaBraM and CBraMod on motor imagery under a validation-locked protocol in which preprocessing, architecture, optimization, freeze depth, checkpoint, temperature, and method selection are determined using training-session data only. On four-class BCI Competition IV-2a, every supervised comparator evaluated here outperforms every foundation-model configuration, including validation-selected fine-tuning. We then examine a key confound: foundation models and task-specific decoders are normally evaluated using different input pipelines. Retraining three supervised architectures on the broadband arrays consumed by the foundation models produces matched-input accuracy differences of opposite sign across architectures: broadband input improves ATCNet by 0.078 accuracy while reducing EEG Conformer accuracy by 0.088. None of the three individual matched-input terms is significant after multiple-comparison correction at n = 9, so we treat the sign variation descriptively rather than as a formal architecture-by-pipeline interaction. These observed sign differences suggest that a single comparator may not provide an architecture-invariant decomposition of a pretrained-versus-supervised performance gap. The four-class deficit also does not reproduce uniformly across motor-imagery datasets: on two-class BNCI2014-004 we cannot detect the same separation between fine-tuned CBraMod and the supervised comparators. Finally, validation-fitted temperature scaling returns foundation-model calibration error to the supervised range despite substantially lower four-class accuracy.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2609.23924",
  "author_total": 2,
  "authors": [
    {
      "name": "Kevin Zhou"
    },
    {
      "name": "Sparsh Roy"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/6cb6yq0bq9xym.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2609.23924",
  "fulltext_url": null,
  "id": "6cb6yq0bq9xym",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2609.23924",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "q-bio.NC"
    ],
    [
      "openalex",
      "2805|EEG and Brain-Computer Interfaces"
    ]
  ],
  "openalex": "W7214049443",
  "pdf_url": "https://arxiv.org/pdf/2609.23924",
  "pmcid": null,
  "pmid": null,
  "published": "2026-09-20",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/6cb6yq0bq9xym.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:23:28Z",
      "native_id": "2609.23924",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2609.23924"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.NC",
      "subfield": "neuroscience"
    }
  ],
  "title": "Matched-Input Estimates Differ in Sign Across Architectures: Auditing EEG Foundation Models on Motor Imagery",
  "title_text": "Matched-Input Estimates Differ in Sign Across Architectures: Auditing EEG Foundation Models on Motor Imagery",
  "type": "preprint",
  "updated_at": "2026-10-30T12:23:28Z",
  "url": "https://research.draconitic.com/paper/6cb6yq0bq9xym",
  "venue": "arXiv",
  "versions": []
}