{
  "schema_version": "bci-report-nsp-components-update-v1",
  "release_id": "nsp-components-update-20261106",
  "generated_at": "2026-11-06",
  "status": "aggregate_preview",
  "publication": "released",
  "title": "A separate panel: the components of Neural State Prediction, tested locally",
  "question": "Do the target components Neural State Prediction proposes change decoding in a small model trained here, and how much of a frozen encoder’s drop under zeroed channels returns when its head is refitted?",
  "placement": {
    "page": "/topics/pilot-studies/",
    "anchor": "nsp-components",
    "after_anchor": "eegmat-ten-second",
    "blocks": [
      {
        "id": "c",
        "anchor": "nsp-components-c"
      },
      {
        "id": "supplement",
        "anchor": "nsp-components-supplement"
      },
      {
        "id": "a",
        "anchor": "nsp-components-a"
      }
    ],
    "pointers_from": [
      {
        "page": "/topics/does-pretraining-help/",
        "block": "supplement"
      },
      {
        "page": "/methods/eeg-fm-masking/",
        "block": "a"
      },
      {
        "page": "/topics/fewer-electrodes/",
        "block": "a"
      }
    ],
    "pointers_carry_figures": false,
    "counted_in_pilots": false,
    "leaderboard": false,
    "core_matrix_row": false,
    "new_dataset_count": 0,
    "new_model_count": 0,
    "text": "A panel of its own on the pilot studies page, after the counted pilots, in three blocks: the local component model, the trained model against the same model never trained, and channels zeroed on two published encoders and a log band-power ridge; the pointers from three pages carry no figures."
  },
  "metric_units": "balanced accuracy and probe accuracy are in percent; paired differences and their interval bounds are in percentage points; content-dependence drops are differences of mean cosine similarity between a model’s predictions and its own fixed Target, without a unit; counts of people, windows, channels, jobs, steps, heads, predictions and draws are whole numbers",
  "scope": "One small model trained here on EEGMAT with each of the two NSP target components on or off, the same model never trained, and two published frozen eeg-fm-masking encoders with a log band-power ridge under zeroed channels. Every score is an equal-person mean on people held out whole; each difference carries a descriptive paired-participant bootstrap interval.",
  "paper": {
    "name": "Neural State Prediction (NSP)",
    "title": "Neural State Prediction: Obstructing Shortcut Learning in EEG Foundation Models",
    "arxiv_id": "2609.31167",
    "version": "v1",
    "url": "https://arxiv.org/abs/2609.31167v1",
    "licence": "CC BY-NC-ND 4.0",
    "licence_url": "https://creativecommons.org/licenses/by-nc-nd/4.0/",
    "cited_only": true,
    "reproduced": false,
    "setting_as_stated": {
      "parameters": 42830000,
      "pretraining_corpus": "TUEG",
      "downstream_datasets": 30
    },
    "components": [
      {
        "id": "resid",
        "label": "Target residualisation (ResID)",
        "definition": "Subtracts the additive channel and time means from the prediction target; every cell keeps the same feature standardisation."
      },
      {
        "id": "topology",
        "label": "Topology-separated context",
        "definition": "Hides the target’s nearest channels and neighbouring time steps from the visible context."
      }
    ],
    "author_checkpoint_found": false,
    "reference_code": {
      "licence": "MIT",
      "linked": false
    }
  },
  "c": {
    "model": {
      "layers": 4,
      "width": 128,
      "parameters": 1378105,
      "steps_per_job": 500,
      "cells": 4,
      "folds": 5,
      "seeds": 3,
      "jobs": 60,
      "readout": {
        "head": "ridge",
        "alpha": 100,
        "read_at": "final step"
      }
    },
    "cohort": {
      "dataset": "EEGMAT",
      "people": 36,
      "windows": 2160,
      "window_seconds": 2,
      "channels": 19,
      "native_hz": 500,
      "sampling_hz": 200
    },
    "topology_rule": "Topology follows the reference code’s rule: a spatial expansion to each target channel and its two nearest neighbours, then a temporal expansion of one step either side, cross terms included; the random control hides as many other positions chosen at random.",
    "cells": [
      {
        "id": "ResID0_Random",
        "resid": false,
        "geometry": "random",
        "balanced_accuracy_pct": 58.719135802469125,
        "people": 36
      },
      {
        "id": "ResID1_Random",
        "resid": true,
        "geometry": "random",
        "balanced_accuracy_pct": 58.44135802469135,
        "people": 36
      },
      {
        "id": "ResID0_Topology",
        "resid": false,
        "geometry": "topology",
        "balanced_accuracy_pct": 59.12037037037037,
        "people": 36
      },
      {
        "id": "ResID1_Topology",
        "resid": true,
        "geometry": "topology",
        "balanced_accuracy_pct": 58.78086419753087,
        "people": 36
      }
    ],
    "contrasts": [
      {
        "id": "ResID1_Topology-minus-ResID1_Random",
        "role": "primary",
        "label": "With ResID on: topology minus random",
        "difference_pp": 0.339506172839508,
        "interval_95_pp": [
          -0.16975308641974923,
          0.8641975308641985
        ],
        "interval_includes_zero": true,
        "reaches_practical_margin": false
      },
      {
        "id": "ResID1_Topology-minus-ResID0_Topology",
        "role": "secondary",
        "label": "With topology: ResID on minus off",
        "difference_pp": -0.3395061728395046,
        "interval_95_pp": [
          -1.2345679012345656,
          0.601851851851853
        ],
        "interval_includes_zero": true,
        "reaches_practical_margin": false
      },
      {
        "id": "ResID-by-geometry-interaction",
        "role": "secondary",
        "label": "ResID × geometry interaction",
        "difference_pp": -0.06172839506172604,
        "interval_95_pp": [
          -0.7098765432098758,
          0.6944444444444463
        ],
        "interval_includes_zero": true,
        "reaches_practical_margin": false
      }
    ],
    "practical_margin_pp": 2,
    "interval": {
      "kind": "descriptive paired participant bootstrap",
      "draws": 10000,
      "p_values": false,
      "multiplicity_correction": false,
      "conditional_on": "the fixed out-of-fold predictions, the overlapping training folds and the paired seeds"
    },
    "probes": {
      "jobs_per_cell": 15,
      "chance_pct": {
        "channel": 5.263157894736842,
        "time": 10.0
      },
      "cells": [
        {
          "id": "ResID0_Random",
          "target_channel_pct": 76.74479166666667,
          "context_channel_pct": 65.90983072916666,
          "target_time_pct": 10.56640625,
          "context_time_pct": 10.431315104166666
        },
        {
          "id": "ResID1_Random",
          "target_channel_pct": 8.382161458333332,
          "context_channel_pct": 62.403971354166664,
          "target_time_pct": 10.305989583333334,
          "context_time_pct": 10.193684895833334
        },
        {
          "id": "ResID0_Topology",
          "target_channel_pct": 77.11751302083333,
          "context_channel_pct": 66.70572916666667,
          "target_time_pct": 10.509440104166666,
          "context_time_pct": 10.423177083333334
        },
        {
          "id": "ResID1_Topology",
          "target_channel_pct": 8.444010416666666,
          "context_channel_pct": 62.537434895833336,
          "target_time_pct": 10.271809895833334,
          "context_time_pct": 10.231119791666666
        }
      ]
    },
    "content_dependence": {
      "within_model_only": true,
      "weighting": "Held-out test people only: within each seed the folds are weighted by their test people, then the paired seeds are averaged.",
      "cells": [
        {
          "id": "ResID0_Random",
          "native_minus_zero_signal": 0.354566797702487,
          "native_minus_different_subject": 0.38451663195083496
        },
        {
          "id": "ResID1_Random",
          "native_minus_zero_signal": 0.5716385577235079,
          "native_minus_different_subject": 0.585844080475369
        },
        {
          "id": "ResID0_Topology",
          "native_minus_zero_signal": 0.3062372728561065,
          "native_minus_different_subject": 0.31690870294853984
        },
        {
          "id": "ResID1_Topology",
          "native_minus_zero_signal": 0.5019289593742079,
          "native_minus_different_subject": 0.5138749519098307
        }
      ]
    },
    "audit": {
      "first_audit": {
        "status": "incomplete",
        "kept": true,
        "cause": "The auditor’s own reference rounded a training-only statistic in single precision."
      },
      "corrected_audit": {
        "status": "passed",
        "jobs": 60,
        "task_heads": 60,
        "diagnostic_heads": 300,
        "predictions_replayed": 25920,
        "training_steps": 30000,
        "encoder_recomputed": false,
        "optimizer_recomputed": false,
        "tolerance_relaxed": false,
        "producer_unchanged": true,
        "aggregate_unchanged": true
      },
      "text": "The independent numeric audit replayed 60 task heads, 300 diagnostic heads, 25,920 predictions and 30,000 training steps; it did not recompute encoder outputs or optimizer updates. The first audit of this block stopped as incomplete because the auditor’s own reference rounded a training-only statistic in single precision; it is kept. A corrected auditor, checked independently, then passed all 60 jobs in double precision with the same tolerances; the producer and the aggregate are unchanged."
    },
    "limitations": [
      {
        "id": "scale",
        "text": "A local model of 1,378,105 parameters, trained on one cohort for 500 steps per job: not the authors’ model, weights or setting (about 42.83M parameters pretrained on TUEG and evaluated on 30 downstream datasets), so it neither replicates nor refutes NSP."
      },
      {
        "id": "intervals",
        "text": "Intervals are descriptive paired-participant bootstraps (10,000 draws) over overlapping training folds and 3 seeds, with no p-value; the 60 jobs are not 60 independent samples."
      },
      {
        "id": "convention",
        "text": "The 2-point line is a convention fixed before the runs, not an equivalence bound."
      },
      {
        "id": "topology-rule",
        "text": "Topology follows the reference code’s rule (a spatial expansion, then a temporal one, cross terms included), not the paper’s simpler union formula."
      },
      {
        "id": "channel-identity",
        "text": "The position probes read channel and time position, not participant identity, and a lower probe of the Target’s channel does not by itself establish a purer neural state."
      },
      {
        "id": "within-model",
        "text": "Each drop compares one model with its own fixed Target, so it is read within a cell, not across cells; the absolute cosines are not comparable between cells and are not published."
      }
    ]
  },
  "supplement": {
    "posthoc": true,
    "budgets_pct": [
      10,
      25,
      100
    ],
    "comparisons_after_main_results": 18,
    "random_init": {
      "definition": "The same architecture from the same initial states as the four trained cells, never trained, read by the same ridge head.",
      "optimizer": false
    },
    "curves": [
      {
        "model": "ResID0_Random",
        "budget_pct": 10,
        "balanced_accuracy_pct": 55.246913580246904
      },
      {
        "model": "ResID0_Random",
        "budget_pct": 25,
        "balanced_accuracy_pct": 56.20370370370371
      },
      {
        "model": "ResID0_Random",
        "budget_pct": 100,
        "balanced_accuracy_pct": 58.719135802469125
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 10,
        "balanced_accuracy_pct": 55.108024691358025
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 25,
        "balanced_accuracy_pct": 55.61728395061727
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 100,
        "balanced_accuracy_pct": 58.44135802469136
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 10,
        "balanced_accuracy_pct": 55.20061728395061
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 25,
        "balanced_accuracy_pct": 56.21913580246913
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 100,
        "balanced_accuracy_pct": 59.12037037037038
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 10,
        "balanced_accuracy_pct": 55.37037037037037
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 25,
        "balanced_accuracy_pct": 55.586419753086425
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 100,
        "balanced_accuracy_pct": 58.78086419753086
      },
      {
        "model": "RandomInit",
        "budget_pct": 10,
        "balanced_accuracy_pct": 53.780864197530875
      },
      {
        "model": "RandomInit",
        "budget_pct": 25,
        "balanced_accuracy_pct": 55.16975308641974
      },
      {
        "model": "RandomInit",
        "budget_pct": 100,
        "balanced_accuracy_pct": 56.62037037037037
      }
    ],
    "trained_minus_random_init": [
      {
        "model": "ResID0_Random",
        "budget_pct": 10,
        "difference_pp": 1.4660493827160492,
        "interval_95_pp": [
          0.30864197530864346,
          2.592592592592593
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID0_Random",
        "budget_pct": 25,
        "difference_pp": 1.0339506172839508,
        "interval_95_pp": [
          -0.2777777777777796,
          2.4537037037037033
        ],
        "interval_includes_zero": true
      },
      {
        "model": "ResID0_Random",
        "budget_pct": 100,
        "difference_pp": 2.098765432098766,
        "interval_95_pp": [
          0.3703703703703706,
          3.935570987654316
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 10,
        "difference_pp": 1.3271604938271606,
        "interval_95_pp": [
          0.09259259259259198,
          2.5154320987654324
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 25,
        "difference_pp": 0.4475308641975311,
        "interval_95_pp": [
          -0.848765432098765,
          1.8364197530864195
        ],
        "interval_includes_zero": true
      },
      {
        "model": "ResID1_Random",
        "budget_pct": 100,
        "difference_pp": 1.8209876543209864,
        "interval_95_pp": [
          -0.077160493827161,
          3.67322530864197
        ],
        "interval_includes_zero": true
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 10,
        "difference_pp": 1.4197530864197536,
        "interval_95_pp": [
          0.2006172839506196,
          2.6080246913580254
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 25,
        "difference_pp": 1.0493827160493825,
        "interval_95_pp": [
          -0.2623456790123468,
          2.4074074074074066
        ],
        "interval_includes_zero": true
      },
      {
        "model": "ResID0_Topology",
        "budget_pct": 100,
        "difference_pp": 2.4999999999999996,
        "interval_95_pp": [
          0.8024691358024695,
          4.243827160493827
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 10,
        "difference_pp": 1.5895061728395055,
        "interval_95_pp": [
          0.36998456790123285,
          2.824074074074074
        ],
        "interval_includes_zero": false
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 25,
        "difference_pp": 0.4166666666666661,
        "interval_95_pp": [
          -0.9413580246913583,
          1.8518518518518523
        ],
        "interval_includes_zero": true
      },
      {
        "model": "ResID1_Topology",
        "budget_pct": 100,
        "difference_pp": 2.1604938271604954,
        "interval_95_pp": [
          0.18518518518518712,
          4.151234567901235
        ],
        "interval_includes_zero": false
      }
    ],
    "mechanism": [
      {
        "id": "ResID1_Topology-minus-ResID1_Random",
        "budget_pct": 10,
        "difference_pp": 0.2623456790123455,
        "interval_95_pp": [
          -0.1851851851851862,
          0.7253086419753074
        ],
        "interval_includes_zero": true
      },
      {
        "id": "ResID1_Topology-minus-ResID0_Topology",
        "budget_pct": 10,
        "difference_pp": 0.1697530864197521,
        "interval_95_pp": [
          -0.7407407407407406,
          1.0648148148148129
        ],
        "interval_includes_zero": true
      },
      {
        "id": "ResID-by-geometry-interaction",
        "budget_pct": 10,
        "difference_pp": 0.30864197530864135,
        "interval_95_pp": [
          -0.3240740740740746,
          0.9413580246913568
        ],
        "interval_includes_zero": true
      },
      {
        "id": "ResID1_Topology-minus-ResID1_Random",
        "budget_pct": 25,
        "difference_pp": -0.03086419753086485,
        "interval_95_pp": [
          -0.4320987654321,
          0.3703703703703682
        ],
        "interval_includes_zero": true
      },
      {
        "id": "ResID1_Topology-minus-ResID0_Topology",
        "budget_pct": 25,
        "difference_pp": -0.6327160493827164,
        "interval_95_pp": [
          -1.5277777777777777,
          0.2932098765432096
        ],
        "interval_includes_zero": true
      },
      {
        "id": "ResID-by-geometry-interaction",
        "budget_pct": 25,
        "difference_pp": -0.04629629629629675,
        "interval_95_pp": [
          -0.6635802469135816,
          0.5709876543209873
        ],
        "interval_includes_zero": true
      }
    ],
    "budget_rule": "Each fold fixes one subset of its outer-training windows for each budget, stratified by person and class, nested across budgets and shared by every seed and model; the scaler and the ridge head are fitted on that subset alone.",
    "replay": {
      "checkpoints": 4,
      "of_checkpoints": 60,
      "comparisons": 40,
      "maximum_scaled_error": 0,
      "optimizer_replayed": false,
      "task_labels_read": false
    },
    "audit": {
      "status": "passed",
      "new_heads": 165,
      "reused_heads": 60,
      "predictions_replayed": 97200
    },
    "limitations": [
      {
        "id": "after-main-results",
        "text": "These 18 comparisons were made after the main results; their intervals are descriptive, conditional on fixed label subsets, overlapping folds and three seeds."
      },
      {
        "id": "baseline",
        "text": "The baseline is the untrained model at 56.62% with every label, not the 50% chance level."
      },
      {
        "id": "budgets",
        "text": "A fitting-label budget is the share of outer-training windows that fit the scaler and the head; every training person is kept and the full labels set the stratification, so it is not a label-acquisition cost."
      },
      {
        "id": "replay-scope",
        "text": "An independent forward replay of 4 final checkpoints, one per cell, matched all 40 array comparisons with a maximum scaled error of 0; it covers 4 of the 60 checkpoints and no optimizer step, and read no task label."
      }
    ]
  },
  "a": {
    "encoders": [
      {
        "id": "mae",
        "label": "eeg-fm-masking MAE (r = 9 cm, L = 2)",
        "checkpoint": "eeg-fm-masking/mae-r9cm-L2"
      },
      {
        "id": "jepa",
        "label": "eeg-fm-masking JEPA (r = 9 cm, L = 2)",
        "checkpoint": "eeg-fm-masking/jepa-r9cm-L2"
      }
    ],
    "comparator": {
      "id": "bandpower",
      "label": "log band-power ridge (5 bands, visible channels)",
      "definition": "five log band powers per visible channel (1–4, 4–8, 8–13, 13–30 and 30–45 Hz, from a Hann periodogram of each two-second window at 200 Hz) with an indicator for each zeroed channel, read by the same ridge head (alpha 100)"
    },
    "cohorts": [
      {
        "task": "eegmat",
        "dataset": "EEGMAT",
        "protocol": "arithmetic-rest",
        "people": 36,
        "windows": 2160,
        "channels": 19,
        "zeroed": 6
      },
      {
        "task": "mi",
        "dataset": "ds003810",
        "protocol": "mi-rest",
        "people": 10,
        "windows": 1200,
        "channels": 15,
        "zeroed": 5
      }
    ],
    "draws_per_window": 8,
    "chance_pct": 50,
    "heads": {
      "fixed": "The scaler and ridge head trained on clean windows, kept fixed for the zeroed windows.",
      "matched": "A scaler and ridge head refitted on the zeroed training windows of the outer-training people only, each original window’s 8 views weighted equally."
    },
    "cells": [
      {
        "task": "eegmat",
        "model": "mae",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 63.56481481481481
      },
      {
        "task": "eegmat",
        "model": "mae",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 54.311342592592595
      },
      {
        "task": "eegmat",
        "model": "mae",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 64.3113425925926
      },
      {
        "task": "eegmat",
        "model": "mae",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 55.02893518518519
      },
      {
        "task": "eegmat",
        "model": "mae",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 63.987268518518526
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 61.52777777777778
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 54.166666666666664
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 63.95833333333333
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 52.719907407407405
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 64.18981481481481
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 64.9537037037037
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 53.391203703703695
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 64.21296296296296
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 52.88194444444444
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 65.05208333333331
      },
      {
        "task": "mi",
        "model": "mae",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 68.08333333333334
      },
      {
        "task": "mi",
        "model": "mae",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 56.552083333333336
      },
      {
        "task": "mi",
        "model": "mae",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 67.375
      },
      {
        "task": "mi",
        "model": "mae",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 57.54166666666666
      },
      {
        "task": "mi",
        "model": "mae",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 65.92708333333334
      },
      {
        "task": "mi",
        "model": "jepa",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 63.16666666666666
      },
      {
        "task": "mi",
        "model": "jepa",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 53.35416666666666
      },
      {
        "task": "mi",
        "model": "jepa",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 61.0
      },
      {
        "task": "mi",
        "model": "jepa",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 53.94791666666667
      },
      {
        "task": "mi",
        "model": "jepa",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 59.354166666666664
      },
      {
        "task": "mi",
        "model": "bandpower",
        "condition": "clean",
        "head": null,
        "balanced_accuracy_pct": 57.33333333333333
      },
      {
        "task": "mi",
        "model": "bandpower",
        "condition": "near",
        "head": "fixed",
        "balanced_accuracy_pct": 52.666666666666664
      },
      {
        "task": "mi",
        "model": "bandpower",
        "condition": "near",
        "head": "matched",
        "balanced_accuracy_pct": 55.677083333333336
      },
      {
        "task": "mi",
        "model": "bandpower",
        "condition": "random",
        "head": "fixed",
        "balanced_accuracy_pct": 51.885416666666664
      },
      {
        "task": "mi",
        "model": "bandpower",
        "condition": "random",
        "head": "matched",
        "balanced_accuracy_pct": 55.291666666666664
      }
    ],
    "contrasts": [
      {
        "task": "eegmat",
        "model": "mae",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "primary",
        "difference_pp": -0.7175925925925939,
        "interval_95_pp": [
          -1.707320601851854,
          0.2951388888888856
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "mae",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -9.253472222222225,
        "interval_95_pp": [
          -13.072916666666668,
          -5.335503472222227
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "mae",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -8.53587962962963,
        "interval_95_pp": [
          -12.094907407407408,
          -5.046151620370372
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "mae",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 0.32407407407407374,
        "interval_95_pp": [
          -1.2269965277777808,
          1.8981481481481457
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "mae",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": 0.7465277777777772,
        "interval_95_pp": [
          -1.1979166666666654,
          2.7141203703703685
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "mae",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": 0.4224537037037035,
        "interval_95_pp": [
          -1.9733796296296289,
          3.015046296296296
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "primary",
        "difference_pp": 1.4467592592592593,
        "interval_95_pp": [
          0.16782407407407604,
          2.708333333333333
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -7.361111111111114,
        "interval_95_pp": [
          -11.244357638888893,
          -3.0844907407407423
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -8.807870370370374,
        "interval_95_pp": [
          -12.939814814814817,
          -4.2418981481481515
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": -0.23148148148148198,
        "interval_95_pp": [
          -1.8172743055555576,
          1.3022280092592575
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": 2.430555555555553,
        "interval_95_pp": [
          0.23726851851851766,
          4.797743055555551
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "jepa",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": 2.6620370370370354,
        "interval_95_pp": [
          -0.3067129629629653,
          5.868200231481478
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 0.5092592592592592,
        "interval_95_pp": [
          -0.7466724537037046,
          1.7766203703703691
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -11.562500000000004,
        "interval_95_pp": [
          -15.601851851851855,
          -7.291666666666671
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -12.071759259259261,
        "interval_95_pp": [
          -16.036024305555557,
          -7.980324074074078
        ],
        "interval_includes_zero": false
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": -0.8391203703703717,
        "interval_95_pp": [
          -2.2164351851851856,
          0.42245370370370217
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -0.7407407407407418,
        "interval_95_pp": [
          -3.425925925925926,
          2.042824074074072
        ],
        "interval_includes_zero": true
      },
      {
        "task": "eegmat",
        "model": "bandpower",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": 0.09837962962962986,
        "interval_95_pp": [
          -2.8994502314814823,
          3.264033564814812
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": -0.989583333333327,
        "interval_95_pp": [
          -2.3333333333333273,
          0.32291666666667385
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -11.531250000000002,
        "interval_95_pp": [
          -14.395833333333329,
          -8.656250000000004
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -10.541666666666675,
        "interval_95_pp": [
          -12.666666666666673,
          -8.48958333333334
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 1.4479166666666674,
        "interval_95_pp": [
          0.23958333333333415,
          2.6979166666666687
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -0.7083333333333341,
        "interval_95_pp": [
          -2.125,
          0.38541666666666474
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "mae",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -2.1562500000000018,
        "interval_95_pp": [
          -4.010416666666672,
          -0.39583333333332965
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": -0.5937500000000034,
        "interval_95_pp": [
          -1.7708333333333364,
          0.5416666666666604
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -9.812500000000004,
        "interval_95_pp": [
          -14.072916666666668,
          -5.406250000000007
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -9.218750000000002,
        "interval_95_pp": [
          -12.791666666666664,
          -5.57291666666667
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 1.645833333333334,
        "interval_95_pp": [
          -0.6770833333333328,
          4.000000000000002
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -2.166666666666669,
        "interval_95_pp": [
          -4.687500000000002,
          0.625
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "jepa",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -3.8125000000000036,
        "interval_95_pp": [
          -7.250000000000003,
          -0.27057291666667166
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 0.7812499999999971,
        "interval_95_pp": [
          -0.10416666666666963,
          1.677083333333331
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -4.66666666666667,
        "interval_95_pp": [
          -7.29166666666667,
          -2.2187500000000027
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "fixed",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -5.447916666666666,
        "interval_95_pp": [
          -8.25,
          -2.718749999999999
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "matched",
        "contrast": "near-minus-random",
        "role": "secondary",
        "difference_pp": 0.38541666666666863,
        "interval_95_pp": [
          -1.0729166666666619,
          1.8020833333333364
        ],
        "interval_includes_zero": true
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "matched",
        "contrast": "near-minus-clean",
        "role": "secondary",
        "difference_pp": -1.6562499999999996,
        "interval_95_pp": [
          -3.166666666666669,
          -0.09374999999999797
        ],
        "interval_includes_zero": false
      },
      {
        "task": "mi",
        "model": "bandpower",
        "head": "matched",
        "contrast": "random-minus-clean",
        "role": "secondary",
        "difference_pp": -2.0416666666666683,
        "interval_95_pp": [
          -3.947916666666669,
          -0.08333333333333516
        ],
        "interval_includes_zero": false
      }
    ],
    "clean_equals_published": [
      {
        "checkpoint": "eeg-fm-masking/mae-r9cm-L2",
        "protocol": "arithmetic-rest",
        "file": "foundation-models-update.json"
      },
      {
        "checkpoint": "eeg-fm-masking/mae-r9cm-L2",
        "protocol": "mi-rest",
        "file": "foundation-models-update.json"
      },
      {
        "checkpoint": "eeg-fm-masking/jepa-r9cm-L2",
        "protocol": "mi-rest",
        "file": "foundation-models-update.json"
      }
    ],
    "clean_differs_from_published": [
      {
        "checkpoint": "eeg-fm-masking/jepa-r9cm-L2",
        "protocol": "arithmetic-rest",
        "file": "foundation-models-update.json"
      }
    ],
    "audit": {
      "status": "passed",
      "heads": 90,
      "predictions_replayed": 332640,
      "encoder_recomputed": false
    },
    "limitations": [
      {
        "id": "not-a-montage",
        "text": "Channels are zeroed after preprocessing, at each model’s input: not a physical montage."
      },
      {
        "id": "refit-competing",
        "text": "Readout mismatch is a competing explanation for the fixed head’s drop, not a unique causal decomposition; the recovery does not show that the zeroed channels carried no task signal."
      },
      {
        "id": "directions-differ",
        "text": "Near-minus-random directions differ between MAE and JEPA."
      },
      {
        "id": "mi-small",
        "text": "Motor imagery against rest has 10 people, and its band-power ridge starts at 57.33%, near the 50% chance level."
      },
      {
        "id": "comparator-recipe",
        "text": "The log band-power ridge is a different recipe from the core Spectral ridge row."
      },
      {
        "id": "not-nsp-weights",
        "text": "MAE and JEPA here are eeg-fm-masking’s published weights, not NSP weights. On all channels, MAE’s two scores and JEPA’s motor-imagery score equal the published frozen-probe cells of these checkpoints; JEPA’s EEGMAT score differs from its published cell."
      }
    ]
  },
  "not_run": [
    {
      "id": "stage-b",
      "text": "The paper’s own encoder: no author checkpoint was found in the sources checked (the paper, its reference code, and model and code searches)."
    },
    {
      "id": "controls",
      "text": "The discriminating controls the design names: shuffled coordinates and distance-matched random context; normalisation-only, channel-only and time-only ResID; batch-partner and global-statistics controls; spectrum- and power-matched mismatch; regulariser controls; a subject-identity probe; longer or larger training and other cohorts."
    }
  ],
  "shared_limitations": [
    {
      "id": "eegmat-order",
      "text": "EEGMAT’s recording order (rest before arithmetic), the source’s prior ICA cleaning and EEGMAT’s use as a development cohort are stated in the NeuroLM panel and bind all three blocks."
    },
    {
      "id": "audit-scope",
      "text": "The numeric audits replayed heads, predictions and summaries from sealed features; they did not rerun the encoders or the training, and the historical raw recordings were not re-hashed."
    }
  ],
  "rights": {
    "datasets": {
      "eegmat": {
        "name": "EEGMAT",
        "source": "https://physionet.org/content/eegmat/1.0.0/",
        "version": "PhysioNet EEG During Mental Arithmetic Tasks 1.0.0 · 1.0.0",
        "license": "Open Data Commons Attribution License 1.0",
        "licenseUrl": "https://opendatacommons.org/licenses/by/1-0/",
        "attribution": "Igor Zyma, Ivan Seleznov, Anton Popov, Mariia Chernykh, Oleksii Shpenkov · EEG During Mental Arithmetic Tasks 1.0.0, PhysioNet, doi:10.13026/C2JQ1P. Study: Zyma et al. (2019), doi:10.3390/data4010014. PhysioNet platform: Pollard et al. (2026), doi:10.1038/s44360-026-00096-z.",
        "privacyReview": "Paper reports Bioethics Commission approval (15 August 2018) and written consent from every participant. EDF dates are normalized, but subject-info.csv contains age, gender, occupation, and recording date; those fields must not be published.",
        "reviewedAt": "2026-09-20",
        "reviewBasis": [
          "https://physionet.org/content/eegmat/1.0.0/",
          "https://www.mdpi.com/2306-5729/4/1/14",
          "https://opendatacommons.org/licenses/by/1-0/"
        ],
        "rightsScope": "Personal noncommercial research; aggregate results only",
        "published_here": "Reused unchanged from the released record of the same recordings. Published here for the NSP component panel: equal-person balanced accuracies, paired differences with their participant-bootstrap intervals, probe and cosine-drop means over jobs, and cohort counts; no per-person, per-fold, per-seed or per-window value."
      },
      "ds003810": {
        "name": "ds003810",
        "source": "https://doi.org/10.18112/openneuro.ds003810.v2.0.2",
        "version": "OpenNeuro snapshot 2.0.2 · c674303319303b9ffc4ff7e2340b3a4807e8ffb5",
        "license": "CC0-1.0",
        "licenseUrl": "https://creativecommons.org/publicdomain/zero/1.0/",
        "attribution": "Peterson et al. · OpenNeuro ds003810, version 2.0.2. Study: https://pmc.ncbi.nlm.nih.gov/articles/PMC9114495/",
        "privacyReview": "Study paper reports ethics approval by CEySTE CCT-CONICET Santa Fe, Declaration of Helsinki compliance, and signed consent. Twelve adults participated; ten were retained. Public participants.tsv includes exact age, sex, handedness, and pseudonymous ID, which must not be surfaced.",
        "reviewedAt": "2026-09-20",
        "reviewBasis": [
          "https://doi.org/10.18112/openneuro.ds003810.v2.0.2",
          "https://pmc.ncbi.nlm.nih.gov/articles/PMC9114495/"
        ],
        "rightsScope": "Personal noncommercial research; aggregate results only",
        "published_here": "Reused unchanged from the released record of the same recordings. Published here for block 3 of the NSP component panel: equal-person balanced accuracies, paired differences with their participant-bootstrap intervals, and cohort counts; no per-person, per-fold or per-window value."
      }
    },
    "models": {
      "eeg-fm-masking": {
        "paper": "https://arxiv.org/abs/2609.33487",
        "arxiv_id": "2609.33487",
        "weights_licence": "CC-BY-4.0",
        "weights_note": "cite arXiv:2609.33487",
        "code_licence": "MIT",
        "code_repository": "https://github.com/PierreGtch/eeg-fm-masking",
        "code_revision": "38a089e63d019c23793201fff58a27d7ce399c1a",
        "weights_distributed": false
      }
    },
    "terms": {
      "record": "https://arxiv.org/abs/2609.31167v1",
      "paper": "CC BY-NC-ND 4.0",
      "code": "MIT (reference code, not linked)",
      "weights": "none found in the sources checked"
    },
    "summary": "EEGMAT keeps its released record: Open Data Commons Attribution License 1.0, aggregate results only. ds003810 keeps its released record: CC0-1.0, aggregate results only. eeg-fm-masking’s weights are CC BY 4.0 (cite arXiv:2609.33487) and its code is MIT. The NSP paper, arXiv:2609.31167v1, is CC BY-NC-ND 4.0 and is cited only; nothing from it is reproduced. The local model follows the paper’s MIT-licensed reference code. No weights are distributed, and no endorsement by the authors of either paper is implied."
  },
  "release_record": {
    "date": "2026-11-06",
    "decided": "2026-11-05",
    "decision": "Publish the NSP component study as its audited aggregates: a panel of its own on the pilot studies page, after the counted pilots, in three blocks (the local component model, the trained model against the same model never trained, and channels zeroed on two published eeg-fm-masking encoders and a log band-power ridge), with the paper cited as arXiv:2609.31167v1 only.",
    "rights": "EEGMAT keeps its released record: Open Data Commons Attribution License 1.0, aggregate results only. ds003810 keeps its released record: CC0-1.0, aggregate results only. eeg-fm-masking’s weights are CC BY 4.0 (cite arXiv:2609.33487) and its code is MIT. The NSP paper, arXiv:2609.31167v1, is CC BY-NC-ND 4.0 and is cited only; nothing from it is reproduced. The local model follows the paper’s MIT-licensed reference code. No weights are distributed, and no endorsement by the authors of either paper is implied.",
    "research_release": "The study’s records mark it complete, independently audited and checked for interpretation; they do not themselves authorise publication. Publication on the site rests on the site owner’s decision of 2026-11-05.",
    "binding": "This record approves the served file the export audit names."
  },
  "not_published": [
    "Per-person, per-fold, per-seed and per-window values of any kind: fold and seed-by-fold results, each job’s diagnostics and their spread, participant identifiers, the selected label rows of the fitting-label subsets, features, fitted heads, predictions, checkpoints and the sealed files. Only equal-person means, paired differences with their descriptive intervals, probe and cosine-drop means over jobs, and counts are published.",
    "Any figure or text of the NSP paper, which is cited only.",
    "Compute, hosts, storage, paths, run times and memory.",
    "Checkpoints and weights: none is distributed; each SHA-256 identifies the file used and grants nothing."
  ],
  "provenance": {
    "manifest_sha256": "738420654f2715c0f53fe629af6fb7ccb6fde0fa6188c6340ee8d92cbbdb010d",
    "source_sha256": {
      "a": "7edd2c521b49d70b01dd714f8f9afdd40db4415647850f02fdd1e853d4aeed8b",
      "c": "e6f2bdc5e739686eaf172361dc71e295b25b596d46efc321c1510c66471886ea",
      "supplement": "34d19a2f8f5055e219745e023243db3aae2a58eaf1b45c4ef033a3876dc9fae9"
    },
    "audit_sha256": {
      "a": "2884cb2dbc755e61d11c1b35135536bc0a72c7b7066fd639fffb6d32032da014",
      "c_first": "711b35eed417c2f231f8f93d0bb294dcce18d631b6b0d6d1beb5a0648dd8aebc",
      "c": "b96afc39d2a5568f82eae275576a6ea59bc79b07473a1bbcc90393ccfa4e7f5e",
      "supplement": "2ff320846af058431dc0ff6bb63bc06f10493320e475263a978dec68e570afc0",
      "replay": "775f23913bbcd7c6b1bc1477d254a88ceabf17552b4244927a27d2393b40d62f"
    },
    "interpretation_sha256": {
      "a": "6b7b2ed1c0845c2592aae5cd41106548d7a4f242d6fce127bb9aee3a07bb8c0e",
      "c": "a1c8d3403ed55fce572fbdefe42c8c79705acd4deaa336b60b6656bd8aba61d9",
      "c_auditor_correction": "c83468ba35bfd588bd7da3ad0bab5b4110067c795c855f7132b29b7a9f61e63a",
      "final": "6125af83c35488572db7831854d09a38ed93b3ad5a606709e72ced1cb0e739bf",
      "supplement": "f32b889b284c6f0b503545ee3a2454b628803f3f580ab439b52a0b354bf37340",
      "supplement_final": "f22c0113d8a949c1d844d9b3742038ae354b19ed5e93bceee919c3583578d7eb",
      "replay": "5ba020f6a6aedcfc753814d53d2069f1ab85f554f0444eb1f872eb62a7ac5a63"
    },
    "protocol_sha256": {
      "main": "207fa86497114fa2531cb02f799b9ab0c919b7e237825e35964981c158b211ee",
      "supplement": "6ddbbbee9194b2ab331a1bcc17717a0b65d780d19179404be602a3816e388b87"
    },
    "checkpoints_as_run": {
      "eeg-fm-masking/mae-r9cm-L2": {
        "repository": "https://huggingface.co/PierreGtch/eeg-fm-masking_mae_r9cm_L2",
        "revision": "c861ade23dddcb0fc0be718944334a2316d635a5",
        "weights_sha256": "d889abd87a245b92f782246f6fd51b23120208f307bc65b4de587a76ea5ed4f2"
      },
      "eeg-fm-masking/jepa-r9cm-L2": {
        "repository": "https://huggingface.co/PierreGtch/eeg-fm-masking_jepa_r9cm_L2",
        "revision": "88877d4cf2304cd3419181a7b1b416ce1f5d508f",
        "weights_sha256": "e9fb1589ddc701017b7bb84035142552b9ee7d512f1dc3267fe24bc1978313b4"
      }
    },
    "released_files_unchanged": [
      "mvp.json",
      "foundation-models-update.json",
      "pilot-studies-update.json",
      "neurolm-eegmat-update.json",
      "neurorvq-update.json",
      "eegmat-ten-second-update.json"
    ]
  },
  "rounding": {
    "rule": "exact value rounded half away from zero at the printed precision",
    "helper_sha256": "1aaf37e466d56c55d5737233f6f989276119e4f9da351110251bb026cb8f4c9b",
    "ties": [
      {
        "leaf": "$.a.cells[17].balanced_accuracy_pct",
        "value": 67.375,
        "fmt": "pct2raw",
        "prints": "67.38%",
        "unsnapped": 67.37499999999999
      },
      {
        "leaf": "$.a.contrasts[22].interval_95_pp[0]",
        "value": -2.125,
        "fmt": "sgr2",
        "prints": "−2.13",
        "unsnapped": -2.125000000000002
      },
      {
        "leaf": "$.a.contrasts[28].interval_95_pp[1]",
        "value": 0.625,
        "fmt": "sgr2",
        "prints": "+0.63",
        "unsnapped": 0.6249999999999936
      }
    ]
  }
}