Core-matrix protocol · Transfer to a new person

Motor imagery & rest on ds003810

The released protocol, with the score of every method run under it.

The protocol at a glance

Task
Rest versus right-hand imagery
Evaluation
Transfer to a new person
Dataset
ds003810
People
10
Data
1,200 epochs · 5 participant-disjoint folds
Input
15 channels · 2-second windows
Chance level
50.0%
Metrics
Balanced accuracy (primary) · Macro F1 (secondary)

Results

Compare within this protocol only: other protocols differ in cohort, electrodes, window or chance level.

Spectral ridge
54.9% (53.2%–56.7%)
CSP+LDA
47.1% (43.8%–50.2%)
LaBraM
53.5% (51.4%–55.4%)
CBraMod
62.9% (59.6%–66.5%)
EEGNet
71.0% (65.0%–76.3%)
TS+LR
58.9% (54.9%–63.9%)
MDM
53.8% (50.8%–57.9%)
ShallowFBCSPNet
68.8% (63.4%–73.9%)
Deep4Net
68.6% (63.3%–74.0%)
Balanced accuracy, every method under this protocol. Dot: the estimate; line: descriptive 95% interval; dashed line: chance level.
MethodTraining modeBalanced accuracyMacro F1ChannelsPeopleScoring time
Spectral ridgeClassical methodSupervised fit54.9%53.2%–56.7%0.544Mean across held-out participants15100.2 s
CSP+LDAClassical methodSupervised fit47.1%43.8%–50.2%At or below chance level0.406Mean across held-out participants15108.8 s
LaBraMFoundation modelFrozen encoder + ridge head53.5%51.4%–55.4%0.509Mean across held-out participants15102.2 s
CBraModFoundation modelFrozen encoder + ridge head62.9%59.6%–66.5%0.617Mean across held-out participants15105.0 s
EEGNetCompact modelScratch · 20 epochs71.0%65.0%–76.3%Single seed — the highest of the seeds run (see Stability)0.680Mean across held-out participants151055.5 s
TS+LRClassical method · OAS · 4–40 Hz · primary row · Local Ubuntu / CPUSupervised fit58.9%54.9%–63.9%0.534Mean across held-out participants1510—
MDMClassical method · OAS · 4–40 Hz · secondary row · Local Ubuntu / CPUSupervised fit53.8%50.8%–57.9%0.420Mean across held-out participants1510—
ShallowFBCSPNetCompact model · matrix recipe · AdamW · Apple M5 / MPSScratch · 20 epochs68.8%63.4%–73.9%0.672Mean across held-out participants1510—
Deep4NetCompact model · matrix recipe · AdamW · Apple M5 / MPSScratch · 20 epochs68.6%63.3%–74.0%0.659Mean across held-out participants1510—

Scoring time covers fitting and prediction in this configuration, may include accelerator waiting, and excludes data preparation. Scoring time is not recorded for the rows of core-matrix-update-20261106.

Not applicable under this protocol: Standard CCA, FBCCA (they need the stimulation frequencies, and this protocol has none); Temporal ridge, XDAWNCov+TS+LR, xDAWN+LDA (they model the waveform of a stimulus-locked transient potential, and this protocol's classes are not defined by one).

A frozen NeuroRVQ probe with its own input chain is on the pilot studies page →

Pick-two protocol comparison

Compare any two methods on Motor imagery & rest

Inspect point estimates and 95% intervals side by side on a shared axis. Paired differences appear only where a paired interval is published. Dropdowns are grouped by family alphabetically — never ordered by score.

0%25%50%75%100%Chance 50.0%ShallowFBCSPNet68.8% [63.4–73.9]EEGNet71.0% [65.0–76.3]
Published paired contrastShallowFBCSPNet − EEGNet: −2.2 pp[−8.0, +3.4 pp]

95% interval, unadjusted · n = 10 participants · Paired interval includes zero

against the matrix's one-seed EEGNet cell; training variance not included

Default view (Motor imagery & rest): ShallowFBCSPNet balanced accuracy 68.8% (95% interval 63.4–73.9%); EEGNet 71.0% (95% interval 65.0–76.3%); chance 50.0%.

Published paired contrasts table (16 pairs)
Pair (left minus right)Mean differenceGoverning intervalParticipantsAdjustment & notes
TS+LR − Spectral ridge+4.0 pp[−1.1, +11.5 pp]1098.75% interval, Bonferroni-adjusted for four contrasts
TS+LR − EEGNet−12.1 pp[−18.1, −5.4 pp]10nominal 95% intervalagainst the matrix's one-seed EEGNet cell; training variance not included
MDM − EEGNet−17.2 pp[−23.7, −10.6 pp]10nominal 95% intervalagainst the matrix's one-seed EEGNet cell; training variance not included
MDM − Spectral ridge−1.1 pp[−4.7, +3.8 pp]10nominal 95% interval
TS+LR − CSP+LDA+11.8 pp[+7.9, +16.8 pp]10nominal 95% interval
MDM − CSP+LDA+6.8 pp[+3.3, +10.8 pp]10nominal 95% interval
TS+LR − LaBraM+5.4 pp[+0.3, +11.3 pp]10nominal 95% interval
TS+LR − CBraMod−4.0 pp[−8.4, +0.9 pp]10nominal 95% interval
TS+LR − MDM+5.1 pp[+2.3, +7.8 pp]10nominal 95% interval
Deep4Net − EEGNet−2.4 pp[−6.8, +2.3 pp]1095% interval, unadjustedagainst the matrix's one-seed EEGNet cell; training variance not included
ShallowFBCSPNet − EEGNet−2.2 pp[−8.0, +3.4 pp]1095% interval, unadjustedagainst the matrix's one-seed EEGNet cell; training variance not included
ShallowFBCSPNet − Deep4Net+0.3 pp[−6.8, +7.7 pp]1095% interval
Log band-power ridge − Spectral ridge+2.4 pp[−1.2, +5.8 pp]1095% paired participant bootstrap interval, unadjusted
LaBraM at Scale Quark-L − LaBraM at Scale Quark-B−2.3 pp[−4.8, −0.3 pp]1095% paired participant bootstrap interval, unadjusted (21 ladder contrasts)
LaBraM at Scale Quark-H − LaBraM at Scale Quark-L+0.6 pp[−3.1, +3.5 pp]1095% paired participant bootstrap interval, unadjusted (21 ladder contrasts)
LaBraM at Scale Quark-H − LaBraM at Scale Quark-B−1.8 pp[−5.2, +1.0 pp]1095% paired participant bootstrap interval, unadjusted (21 ladder contrasts)

How these rows were run

Riemannian and xDAWN baselines

Recipe
TS+LR: OAS · 4–40 Hz · primary row; MDM: OAS · 4–40 Hz · secondary row
Hardware
TS+LR, MDM: Local Ubuntu / CPU

One fixed configuration per method: library or MOABB-reference settings with the stated deviations. The 4–40 Hz band-pass (or demeaning plus 50 Hz decimation on the ERP protocols) is part of the recipe and is applied to each window on its own.

ShallowFBCSPNet, Deep4Net and CSP+LDA under the matrix recipe

Recipe
ShallowFBCSPNet, Deep4Net: matrix recipe · AdamW
Hardware
ShallowFBCSPNet, Deep4Net: Apple M5 / MPS

Paired differences

Each difference is the mean over people of each person’s difference in balanced accuracy, in percentage points (pp), on the same windows and people.

Riemannian and xDAWN baselines

DifferenceMean (pp)98.75% interval, Bonferroni-adjusted for four contrasts95% interval, unadjustedReading
TS+LR − Spectral ridgeTS+LR · OAS · 4–40 Hz+4.0 pp−1.1 to 11.5−0.3 to 9.7unresolved

Paired participant bootstrap (10,000 draws). The four contrasts against the Spectral ridge cells carry 98.75 % intervals (Bonferroni over four); the other contrasts carry nominal 95 % intervals with no familywise claim. 'Higher' or 'lower' means the interval lies wholly above or below zero; 'unresolved' means it includes zero.

Ten people: a percentile bootstrap over ten people tends to give intervals narrower than their nominal level.

ShallowFBCSPNet, Deep4Net and CSP+LDA under the matrix recipe

against the matrix's one-seed EEGNet cell; training variance not included
DifferenceMean (pp)95% interval, unadjustedBonferroni interval over 18 contrastsReading
Deep4Net − EEGNet−2.4 pp−6.8 to 2.3−8.6 to 5.2unresolved
ShallowFBCSPNet − EEGNet−2.2 pp−8.0 to 3.4−10.8 to 5.7unresolved

The paired 95 % interval governs the wording; the Bonferroni interval beside it covers all 18 contrasts. Read at 95 % without correction, about one in 20 contrasts with no true difference may show one by chance.

Beside the core matrix

Log band-power ridge (5 bands, visible channels)

Five log band powers (1–4, 4–8, 8–13, 13–30 and 30–45 Hz) per channel on all recorded channels of the protocol, from one Hann periodogram of each published window after removing each channel's window mean and a per-window robust amplitude scaling, read by the core ridge head (alpha 100) on the published windows and participant-held-out folds. It uses the same windows, people, folds and ridge head as the Spectral ridge row, repeated from the table above.

MethodTraining modeBalanced accuracyMacro F1ChannelsPeople
Spectral ridgeClassical method · core matrix row aboveSupervised fit54.92%53.17%–56.67%0.5441510
Log band-power ridge (5 bands, visible channels)Classical methodSupervised fit57.33%54.42%–60.33%0.5581510

Differences from two published rows

Each difference is this row minus the named row on the same people, with a paired participant bootstrap (10,000 draws): an interval wholly above zero reads above, wholly below reads below, and one that includes zero reads not separated, which does not show the two are equivalent. The intervals are descriptive, conditional on these folds, with no multiplicity correction.

Minus the Spectral ridge row: +2.42 pp (−1.17 to +5.83): not separated. The two rows differ in bands, normalisation, estimator, frequency range and sampling rate at once, so the difference measures the whole recipe.

Minus SingLEM, the highest of the 14 frozen-encoder rows this comparison was specified against, chosen on these same test folds: −14.67 pp (−19.33 to −9.75): below. Choosing it on the test folds leans the comparison against this row.

Notes on this comparator

The band-power rows in the pilot studies panels on this cohort come from their own input chains and are read on their own.

On these windows and folds this is the same comparator as the third block of the panel on the components of Neural State Prediction on the pilot studies page →

Beside the core matrix

Further foundation encoders, frozen

Each encoder replaces LaBraM or CBraMod in the frozen recipe above; the windows, people, folds, ridge head and scoring stay the same. Rows are grouped by model family; a row is above or below another only where their 95% intervals do not overlap. Each interval is a descriptive participant bootstrap with no multiplicity correction, and with 126 frozen cells an occasional non-overlap is expected by chance. Every encoder sees this protocol’s published windows, usually shorter than its pretraining context.

Spectral ridge
54.9% (53.2%–56.7%)
CSP+LDA
47.1% (43.8%–50.2%)
LaBraM
53.5% (51.4%–55.4%)
CBraMod
62.9% (59.6%–66.5%)
EEGNet
71.0% (65.0%–76.3%)
REVE Base
66.2% (61.8%–70.6%)
REVE Large
70.2% (64.0%–76.1%)
LUNA Base
52.6% (50.2%–54.9%)
LUNA Large
49.9% (46.6%–53.5%)
BrainOmni Base
58.8% (57.0%–60.8%)
CodeBrain
58.8% (54.5%–62.6%)
EEGMamba
67.3% (63.6%–70.6%)
ST-EEGFormer Base
58.1% (53.7%–63.3%)
ST-EEGFormer Large
62.4% (57.4%–67.8%)
eeg-fm-masking MAE (r=9 cm, L=2)
68.1% (64.3%–72.3%)
ERP-FM Base
54.8% (51.8%–57.8%)
SingLEM
72.0% (66.6%–76.6%)
ZUNA 1.1
50.7% (49.2%–52.1%)
Balanced accuracy under this protocol. Grey: the released rows above; rust: the further encoders, in family order. Dot: the estimate; line: 95% interval; dashed line: chance level.
EncoderTraining modeBalanced accuracyMacro F1ChannelsPeoplePretraining exposure
REVE Base1Foundation modelFrozen encoder + ridge head66.2%61.8%–70.6%0.6401510not in the authors' published pretraining list (checked 2026-10-04)source ↗
REVE Large2Foundation modelFrozen encoder + ridge head70.2%64.0%–76.1%0.6731510not in the authors' published pretraining list (checked 2026-10-04)source ↗
LUNA Base3Foundation modelFrozen encoder + ridge head52.6%50.2%–54.9%0.5031510not in the authors' published pretraining list (checked 2026-10-04)source ↗
LUNA Large4Foundation modelFrozen encoder + ridge head49.9%46.6%–53.5%At or below chance level0.4861510not in the authors' published pretraining list (checked 2026-10-04)source ↗
BrainOmni Base5Foundation modelFrozen encoder + ridge head58.8%57.0%–60.8%0.5821510not in the authors' published pretraining list (checked 2026-10-04)source ↗
CodeBrain6Foundation modelFrozen encoder + ridge head58.8%54.5%–62.6%0.5731510not in the authors' published pretraining list (checked 2026-10-04)source ↗
EEGMamba7Foundation modelFrozen encoder + ridge head67.3%63.6%–70.6%0.6691510not in the authors' published pretraining list (checked 2026-10-04)source ↗
ST-EEGFormer Base8Foundation modelFrozen encoder + ridge head58.1%53.7%–63.3%0.5411510not in the authors' published pretraining list (checked 2026-10-04)source ↗
ST-EEGFormer Large9Foundation modelFrozen encoder + ridge head62.4%57.4%–67.8%0.5861510not in the authors' published pretraining list (checked 2026-10-04)source ↗
eeg-fm-masking MAE (r=9 cm, L=2)10Foundation modelFrozen encoder + ridge head68.1%64.3%–72.3%0.6771510not in the authors' published pretraining list (checked 2026-10-04)source ↗
ERP-FM Base11Foundation modelFrozen encoder + ridge head54.8%51.8%–57.8%0.5241510not in the authors' published pretraining list (checked 2026-10-04)source ↗
SingLEM12Foundation modelFrozen encoder + ridge head72.0%66.6%–76.6%0.7151510not in the authors' published pretraining list (checked 2026-10-04)source ↗
ZUNA 1.113Foundation modelModel card: research use only, not for diagnosis or clinical use.Frozen encoder + ridge head50.7%49.2%–52.1%Interval reaches chance level0.4991510Exposure unknownunknown: the authors do not list their pretraining data closely enough to decidesource ↗

Pretraining exposure is what each model’s authors publish as their pretraining list, checked on 2026-10-04: a dataset absent from it is “not in the authors’ published pretraining list”, which is not proof that its recordings were never seen. ST-EEGFormer lists BETA and SingLEM lists TMNRED; ZUNA 1.1 publishes no list, so every one of its cells is unknown; EEGMamba’s list is read from its official code (its paper was not read), with medium confidence.

Row footnotes

  1. Input removes the per-segment per-channel mean before the published /100 scaling (declared before scoring); a no-mean-removal sensitivity run scores P300 0.71 percentage points higher and sleep 1.40 lower.
  2. Input removes the per-segment per-channel mean (declared before scoring); the no-mean-removal sensitivity run scores P300 2.06 percentage points lower and sleep 0.93 lower.
  3. Resampled to 256 Hz and truncated to whole 40-sample patches; 1-2 s windows and 4-8 channel montages are outside the authors' tested regime.
  4. As LUNA Base. Frozen probes only.
  5. p300-target and semantic-target not run: the 2 s tokenizer window cannot be met by 1 s segments without zero padding. Upstream common average reference kept on 4-8 channel montages.
  6. Published CBraMod pooling (patch mean, channels kept); LoRA effectively adapts only the V projection (1-token attention window upstream).
  7. Published CBraMod pooling; LoRA on Mamba2 in_proj and out_proj (24 targets).
  8. BETA cells are pretraining-exposed (BETA is in the ST-EEGFormer corpus).
  9. BETA cells are pretraining-exposed. The authors' downstream class (1-based temporal index) was used; Large may have been pretrained with a 0-based index. Frozen probes only.
  10. Paper-recommended masking geometry; channel-retained pooling (C x 512). Three sibling checkpoints appear in the masking ablation panel.
  11. In design: p300-target and semantic-target. The other six protocols are negative controls outside the model's design. ERP windows have no pre-stimulus baseline.
  12. semantic-target is pretraining-exposed (TMNRED is in the SingLEM corpus). Each channel is encoded independently; the last 0.25 s of each 2 s segment is not seen by the tokenizer.
  13. Pretraining exposure unknown for every core dataset (no published source list). Encoder only; no channel upsampling or reconstruction.
  14. A sibling of the paper-recommended checkpoint in the masking ablation: the same architecture, corpus and recipe, another masking framework or geometry.
Masking ablation: three sibling checkpoints of eeg-fm-masking

Intervals are marginal: the same people under two checkpoints are not tested as a pair.

EncoderTraining modeBalanced accuracyMacro F1ChannelsPeoplePretraining exposure
eeg-fm-masking MAE (r=9 cm, L=2)10Foundation model · matrix row aboveFrozen encoder + ridge head68.1%64.3%–72.3%0.6771510not in the authors' published pretraining list (checked 2026-10-04)source ↗
eeg-fm-masking JEPA (r=9 cm, L=2)14Foundation modelFrozen encoder + ridge head63.2%58.2%–67.9%0.6201510not in the authors' published pretraining list (checked 2026-10-04)source ↗
eeg-fm-masking MAE (r=one channel, L=1)14Foundation modelFrozen encoder + ridge head64.9%60.3%–69.4%0.6451510not in the authors' published pretraining list (checked 2026-10-04)source ↗
eeg-fm-masking JEPA (r=one channel, L=1)14Foundation modelFrozen encoder + ridge head63.3%59.0%–67.4%0.6281510not in the authors' published pretraining list (checked 2026-10-04)source ↗

Weights licences and terms

No model’s authors endorse these results, and no adapted weights or LoRA deltas were kept or shared.

Notes on these rows

Beside the core matrix

EEGPT, frozen

EEGPT ran in the same runs as the published LaBraM and CBraMod rows above, with only the encoder swapped: the same windows, people, folds, ridge head and scoring.

Spectral ridge
54.9% (53.2%–56.7%)
CSP+LDA
47.1% (43.8%–50.2%)
LaBraM
53.5% (51.4%–55.4%)
CBraMod
62.9% (59.6%–66.5%)
EEGNet
71.0% (65.0%–76.3%)
EEGPT
58.3% (54.9%–61.8%)
Balanced accuracy under this protocol. Grey: the released rows above; rust: EEGPT. Dot: the estimate; line: 95% interval; dashed line: chance level.
EncoderTraining modeBalanced accuracyMacro F1ChannelsPeoplePretraining exposure
EEGPTFoundation modelFrozen encoder + ridge head58.3%54.9%–61.8%0.5551510not in the authors' published pretraining list (checked 2026-10-12)source ↗

Footnotes to this row

  1. ds003810 is recorded at 125 Hz; its windows reach the encoder upsampled to 256 Hz, so they carry nothing above 62.5 Hz.
  2. Every core window is shorter than EEGPT’s 4-second pretraining window: 1 or 2 seconds, and sleep as fifteen 2-second segments. The encoder computes each 0.25-second patch the same way at any length; a shorter window gives fewer patches to average.
  3. The readout is the site’s fixed probe on the frozen summary tokens, as for the published LaBraM and CBraMod rows: ridge (alpha 100) on 2,048 pooled values, and on idle a per-person logistic head. It is not the authors’ downstream probe, a learnable spatial filter with a linear head, and the windows are fed without the average reference EEGPT was pretrained with.

Against the published rows

Each row of the table above that experiments.json holds, by whether its 95% interval and EEGPT’s overlap (no multiplicity correction):

EEGPT’s exposure is read from its authors’ published pretraining list, checked on 2026-10-12, as for the further encoders. Its pretraining corpus: Five public datasets, as the paper’s Table 1 lists them: PhysioMI (PhysioNet motor imagery and execution), HGD, TSU (the Tsinghua 40-target SSVEP benchmark), SEED and M3CV; 58 electrodes, 256 Hz, millivolts, average reference, 4-second windows.

Licence and terms

Frozen-probe comparison

LaBraM at Scale and REVE Small, frozen

Six LaBraM at Scale encoders and REVE Small’s encoder in the same frozen recipe as the rows above: the same windows, people, folds, head and scoring, with no random-initialization control.

Balanced accuracy with its 95% interval, and macro F1, frozen encoders.
EncoderBalanced accuracy (95% interval)Macro F1Exposure
LaBraM at Scale · Quark ladder
Quark-BLocal Ubuntu / CPU61.9%59.1%–64.4%0.612—
Quark-LLocal Ubuntu / CPU59.6%56.3%–62.9%0.578—
Quark-HLocal Ubuntu / CPU60.2%57.4%–62.9%0.596—
Quark-G†Local Ubuntu / CUDA58.7%57.3%–60.1%0.579—
LaBraM at Scale · Atom-L and Molecule-L
Atom-LLocal Ubuntu / CPU56.9%53.3%–61.0%0.539its pretraining corpus includes this protocol's dataset
Molecule-LLocal Ubuntu / CPU59.6%57.2%–61.4%0.577its pretraining corpus includes this protocol's dataset
REVE Small · early checkpoint (epoch 15 of 200)
REVE SmallLocal Ubuntu / CPU61.5%58.3%–64.8%0.590—

† Quark-G: Run on Local Ubuntu / CUDA; its step is read against a GPU re-run of Quark-H and includes the probe's input width doubling from 800 to 1600 under one fixed penalty (see the topic page).

Unless a cell is flagged, its dataset is not in that collection’s published pretraining list (the LaBraM at Scale pretraining manifest at faa387d9, checked 2026-11-05); REVE Small’s is read from the REVE corpus list (checked 2026-10-04).

The Quark ladder’s paired differences across protocols, the Quark-G step among them, are on “Does pretraining help EEG foundation models like LaBraM and CBraMod?”.

Terms: LaBraM at Scale’s code and weights under MIT (Pan-Junyu/LaBraM-at-Scale @ a0b0ce27); REVE Small under the REVE Responsible Use License v1.0 (reve-small @ 3565bdcb), aggregate scientific results only. No weights are distributed.

Two more models · this protocol’s people and folds

Signal-JEPA and EEGSimpleConv

Signal-JEPA as a frozen encoder with a ridge head fitted on the training people, and EEGSimpleConv trained from scratch for 20 epochs, both on this protocol’s windows, people and 5 folds. Each is compared with EEGNet on the same people.

ModelTraining modeInputBalanced accuracyMacro F1Difference from EEGNet
Signal-JEPAFoundation model · 3,456,852 parametersFrozen encoder + ridge head0.5–40 Hz band-pass, 128 Hz67.25%63.42%–70.50%0.669−3.75 pp(−7.58 to −0.17)vs this page’s EEGNet cell: 71.0%against the matrix's one-seed EEGNet cell; training variance not included
EEGSimpleConvCompact model · 185,022 parametersTrained from scratch, 20 epochsthe matrix’s 250 Hz windows59.33%mean of three seeds0.530−10.14 ppvs EEGNet’s three-seed mean · no interval

Each difference is a paired comparison with EEGNet on the same people.

Notes on Signal-JEPA and EEGSimpleConv

Signal-JEPA’s input: each window band-passed 0.5–40 Hz and resampled to 128 Hz, the format it was pretrained on, in microvolts, recorded channels only. The further-encoder panel follows its own input contract, so this row sits apart from it.

Signal-JEPA’s difference is paired, on the same people, against this page’s EEGNet cell: 10,000 participant bootstrap draws, no multiplicity correction. EEGNet’s mean over its three seeds here is 69.47%.

EEGSimpleConv: the mean of three seeds, at width 64 and the matrix’s 20-epoch budget, fixed before scoring. Its difference is from EEGNet’s mean over the same three seeds and is printed without an interval.

Signal-JEPA’s model card names Lee2019 (OpenBMI) as its pretraining data.

Terms of Signal-JEPA and EEGSimpleConv

Limits

Ten-person laboratory task with prompted rest. This is not continuous-idle monitoring or a physical low-channel headset test.

The protocol, step by step

  1. 5 participant-disjoint folds. All recordings from a person stay together. Each person contributes to the held-out predictions once.
  2. epoching: two seconds from the class annotation onset; amplitude: converted to microvolts according to source calibration, then per-channel epoch mean removed; resampling: none; native sampling rate retained; selection: natural file/event order; when over the cap, retain 60 evenly spaced event indices per participant/class; source_units: Microv declared by BIDS channels.tsv; EDF physical dimension is absent, so MNE returns the source numeric microvolt values without SI scaling
  3. One fixed seed (20260919); no early stopping or test-based tuning. EEGNet trains for 20 epochs per fold. Frozen encoders use training-only standardized ridge heads (alpha 100).
  4. Labels: rest, right_hand_imagery.
  5. Uniform-guessing reference: 50.00%. Scores weight participants equally. Intervals describe participant variation; cross-validation training sets overlap.
  6. run 0 is real dominant-hand movement and is excluded; only imagery runs 1-4 contribute
  7. the source reports online 0.5-45 Hz filtering
  8. controlled cue-locked laboratory windows; this does not measure continuous false activations
  9. foundation-model pretraining overlap is unknown
  10. EEGNet three-seed mean 69.47%; sample SD 1.43 percentage points; range 68.17–71.00%. Main table retains the original fixed seed; this is not a confidence interval.

Stability

EEGNet three-seed mean 69.47%; sample SD 1.43 percentage points; range 68.17–71.00%. Main table retains the original fixed seed; this is not a confidence interval.

Seeds run for EEGNet: 71.00% · 69.25% · 68.17%; mean 69.47%

Same participants, folds, preprocessing and 20-epoch budget. Three seeds measure initialization variability, not population uncertainty. Main table retains its preselected seed; no best-seed selection.

Pretraining exposure

Unknown unless explicitly documented; no unseen-pretraining claim.

Checked 2026-10-04. LaBraM: not in the authors' published pretraining list (checked 2026-10-04) source ↗; CBraMod: not in the authors' published pretraining list (checked 2026-10-04) source ↗. The released sentence above predates this check. Further encoders: the exposure column of their table above → · EEGPT: its own section above →

Notes on the methods

Model terms

Source and licence

ds003810

Credit Peterson et al. · OpenNeuro ds003810, version 2.0.2. Study: https://pmc.ncbi.nlm.nih.gov/articles/PMC9114495/

The dataset authors ask you to cite: Peterson V, Galván C, Hernández H, Spies R. A feasibility study of a complete low-cost consumer-grade brain-computer interface system. Heliyon 6(3):e03425 (2020). doi ↗ · as their record asks ↗

Licence CC0-1.0 ↗

Dataset record ↗ · Every result on this dataset →

The figures here are aggregate measurements computed by BCI Report under the licence above; the data remain their creators’.

Permitted scope Personal noncommercial research; aggregate results only

Privacy Only cohort aggregates are published here: no recording, no participant identifier, no per-person score. The full review note is in the protocol JSON ↓

Public-data register note Clear CC0 snapshot plus dataset-specific ethics and signed-consent evidence; aggregate-only publication materially limits privacy exposure.

Rights reviewed · against doi.org ↗ · pmc.ncbi.nlm.nih.gov ↗

Correction on record · : The ds003810 credit links only the 2022 Data in Brief description. The OpenNeuro record asks users to cite Peterson, Galván, Hernández and Spies, Heliyon 6(3):e03425 (2020); the dataset page now gives both. Corrections register →

Reproducibility record

Protocol id
parallel-fixed-subject-folds-v1/ds003810
Dataset release
OpenNeuro snapshot 2.0.2 · c674303319303b9ffc4ff7e2340b3a4807e8ffb5
Hardware
Apple M5 / MPS
Scoring stage sum
71.7 s · may include accelerator waiting
Audit record SHA-256
e762cf7f048bf2df47b9399b9df6171733a93c162ce7800560952df30491a7f2
Summary SHA-256
0dd66f4bd4d083fca1d1ab44a2d2951d57f6d19378079b8cb5ff3ee9241f0cda
Protocol SHA-256
57ddb7692e3bb248b95c4e81b6aa02e7c8b19d774ed4a8b799c811b6ef3b65a2

Rows of core-matrix-update-20261106

Riemannian and xDAWN baselines

Protocol SHA-256
4310329a2d6dddaeff64a965750b64237b08220fc23717b1f3d0c1a48f56b055
Audit record SHA-256
ce1ee2a0345358c613aa89298c1cb90939df15d68d6c48298b99f9daa7f69e82
Hardware
Local Ubuntu / CPU
Row intervals
participant bootstrap, 10,000 draws
Paired-difference intervals
paired participant bootstrap, 10,000 draws

ShallowFBCSPNet, Deep4Net and CSP+LDA under the matrix recipe

Protocol SHA-256
a216bfac4ce6dd11c63d499b58bc93aebb89025168911b5d7d7fe265a866da4c
Audit record SHA-256
f5a1c3dab4a7e265fa3413ada48edc89b9fd4ac98ab7dc0d0f9279761e7e0a9e
Hardware
Apple M5 / MPS
Row intervals
participant bootstrap, 10,000 draws
Paired-difference intervals
paired participant bootstrap, 10,000 draws

One training run per cell with the matrix seed and the matrix recipe: AdamW, 20 epochs (sleep 10), no schedule, no early stopping. This is the recipe of the EEGNet cells; it is not tuned for any network and it is not the original papers' training. Intervals describe variation across people, not across training runs.

Downloads

This protocol's rows in core-matrix-update-20261106 · CSV ↓ · Core matrix: rows run under its exact protocols · JSON ↓

Other protocols

All protocols → · The core matrix on the home page → · All datasets → · All methods →

Cite this page

BCI Report (2026). Motor imagery & rest on ds003810. https://bci.report/protocols/mi-rest/

Figures from releases core-matrix-update-20261106 (2026-11-06), logpower-ridge-update-20261106 (2026-11-06), signal-jepa-eegsimpleconv-update-20261106 (2026-11-06), foundation-checkpoints-update-20261106 (2026-11-06), eegpt-core-update-20261013 (2026-10-13), foundation-models-update-20261004 (2026-10-04), research-preview-20260920 (2026-09-20). Cite the upstream dataset as well: its credit is on this page.

Every release is archived on Zenodo: doi:10.5281/zenodo.23123296. BibTeX for the site and its releases → · CITATION.cff ↗