Questions

What the evidence can answer

Each question has its own page: a short answer, the evidence with its cohort and interval, and its limits. Below them, a map of which kinds of transfer have been measured, which are held, and which have not been tried.

Aggregate measurements, organized by the decision they can inform. They are repeated conditions within protocols, not independent experiments, and not an overall ranking.

Transfer

Does a decoder still work when the sensor, the display, the electrode layout or the body’s movement changes?

Sensor transfer

Dry vs. wet electrodes

What changes when a decoder crosses between two native eight-channel recordings from the same 102 people?

2 s SSVEP · 12 targets · balanced accuracyRead the evidence →Context transfer

Screen to VR

The same 21 people calibrated on a PC screen and tested in a VR headset, and the reverse — and why the result is not the cost of the change. Plus: a P300 decoder trained at one image rate and tested at another.

P300 · 21 people in VR · 9 people across image ratesRead the evidence →Montage

Fewer electrodes

In-ear against scalp for sleep, and four posterior electrodes against sixteen for eyes open or closed — and what neither says about any headset.

Two paired comparisons · 10 and 19 peopleRead the evidence →Motion robustness

On the move

Standing, walking and running results, with scalp and ear recordings and incompatible time windows kept apart.

SSVEP balanced accuracy · ERP ROC AUCRead the evidence →

Adapting models

How much calibration a decoder needs, which part of a pretrained model to update, and whether pretraining helps at all.

Calibration budget

How much calibration?

Twelve, 24 or 48 labeled trials on a wearable SSVEP task, and up to 40 from a person’s next motor-imagery session, help some methods more than others and not every person—and trial count is not elapsed time.

SSVEP · 102 people · motor imagery, next session · 51 peopleRead the evidence →Model adaptation

Which part to update?

LaBraM on new people, same task, zero labels from the test person: head only, last block or LoRA, printed beside the core matrix’s frozen readout. Plus: the next-day experiment, run and held.

EEGMAT, 36 people · three update rules · next day: heldRead the evidence →Representation controls

Does pretraining help?

Matched pretrained and constructor-random encoders under fixed and train-selected readout settings.

Two tasks · two encoders · three random initializationsRead the evidence →

Reliability & clinical

When a decoder should not act, what accuracy hides in sleep staging, and what resting-state EEG can and cannot say about a clinical group.

Abstention · Jev-style

When not to act

A decoder that never acts never fires by mistake. Command detection and false activation, read together — and a Jev-style research plan for when to act, wait or recalibrate.

Idle, 4-person pilot · non-control, 20 people · research planRead the evidence →Sleep staging · simple baselines

Sleep-stage balance

Two simple baselines in 25 healthy sleepers and 55 people with sleep apnoea, as two separate experiments: accuracy beside balanced accuracy, and the stage the stager never predicts.

Dreem DOD-H and DOD-O · 25 and 55 nights · balanced accuracyRead the evidence →Clinical research

Clinical groups

A 149-person Parkinson's and control comparison, with an age-and-sex-only comparator printed beside it — and why neither is a diagnosis.

149 people · one site · balanced accuracyRead the evidence →

Transfer coverage

Which kinds of transfer have been measured

Rows are what changes between training and test; columns are the state of the evidence. Each entry links the page that holds it.

What changes between training and test, and whether this site has evidence on it. Cohort sizes and links only, never scores.
What changes	Measured	Status only	Held	Not measured
PersonNew people: trained on others, tested on someone the model never saw	Core matrix · new-person protocolsEvery protocol whose split is “transfer to a new person”: participant-disjoint folds. Dry vs. wet electrodesn=102also changes: sensorSource models trained on other people, so the new person and the new sensor arrive together. LaBraM adaptation · EEGMATn=36New people, same task, zero labels from the test person. Standing to walking and running, SSVEPn=23also changes: movementTrained on other people standing, tested on a new person on the move. Pretraining controlsn=36 / 10Frozen encoders with a readout fitted on other people, on a motor-imagery and a mental-workload task. Parkinson’s disease vs. controlsn=149Every person held out once; one site, and not a diagnosis. Sleep staging · Dreem, two cohortsn=55 / 25Each night held out once and scored by a model fitted on other people’s nights of the same cohort; the two cohorts are separate experiments.	None	Neural and foundation models on the Dreem cohortsNot run: the source’s physical units disagree. Foundation models on the clinical cohortNot run: the source states no physical amplitude unit.	None
Session / dayThe same person, another session or another day	Mobile ERP · first session to the othersn=24 / 17also changes: movementFitted on the person’s first session, standing; the session and the movement change together. OpenBMI motor imagery · first session to the secondn=51The same person and task: trained on the first session, tested on the second, with and without a few of its labelled trials.	Cross-session pilotn=1One person: every score would be that person’s, so none is published.	Next-day adaptationRun and independently replayed; held for a licence and ethics review.	None
Sensor / montageAnother sensor type or another set of electrodes	Dry ⇄ wet electrodesn=102also changes: personTrained on one sensor type, tested on the other, in both directions. In-ear vs. scalp, sleepn=10comparisonBoth electrode sets scored on the same epochs: a paired comparison, not training on one and testing on the other. Posterior subset vs. all electrodesn=19comparisonA software subset of one headset, not two devices.	None	None	From one headset or amplifier to anotherEvery sensor change here is within one study’s own equipment.
ContextAnother display, movement, or image rate	Screen ⇄ VR, P300n=21The same people, calibrated on one display and tested on the other. Standing to walking and running, SSVEPn=23also changes: personTrained on other people standing, tested on a new person on the move. Standing to walking and running, ERPn=24 / 17also changes: sessionThe same person; movement and session change together. Image rate, P300n=9also changes: recordingTrained on one recording at one rate, tested on a different recording: rate and recording change together.	None	None	None
DatasetTrained on one dataset, tested on another	None	None	None	Training on one dataset, testing on anotherNot run here. Whether a dataset was in a model’s pretraining data is a different question.

A state says whether evidence exists, not whether transfer works. n is the number of people behind an entry; two sizes (n=24 / 17) are two analyses or cohorts behind the same entry, largest first. “comparison” marks two electrode sets scored on the same data rather than a transfer from one to the other; “also changes” names what else changed in the same contrast. A held entry is described in the holds register, not here.

Cohort sizes are read from the reviewed downloads: deployment-topics.json · adaptation-update.json · clinical-update.json · large-source-update.json · context-update.json · evidence-update.json · extension-update.json · Full holds register →