Nov. 13, 2026
Preview page published
This page shows where Nasion is headed. The software is not out yet.
Run any tool on public recordings and watch what happens when the brain is taken out.
In a 2022 study of stroke patients, the eye and muscle signals hiding in the EEG predicted movement better than the brain activity did. Nasion is built to catch that trap in any tool you run through it.
McDermott et al., "Artifacts in EEG-Based BCI Therapies: Friend or Foe?", Sensors 22(1):96, 2022.
This page shows where Nasion is headed. The software is not out yet.
We open with the oldest effect in the book, the one Hans Berger found in 1929. Every result after that comes with its own reality checks.
Run our test yourself and see whether you get the same answer. Nobody has to take our word for it.
We feed every tool fake brain signals, then add blinks and muscle twitches to real ones. A tool that still scores on the fakes is reading noise, and one that climbs with the blinks has learned to watch your eyes instead of your brain.
The planned set includes phase-randomized surrogates, pink noise and shuffled labels, plus recorded eye and muscle artifacts added to real data.
MOABB already ranks the best motor imagery pipelines on public recordings. We take those same pipelines and find out how much of each score is still standing once the brain is pulled out of the signal.
Pipelines from MOABB 1.7.2 (NeuroTechX, BSD license). Data is the EEG Motor Movement/Imagery Dataset on PhysioNet, 109 volunteers at 64 channels in EDF+, sampled at 160 Hz except the imagery runs of subjects 88, 92 and 100, which are at 128 Hz and are resampled to 160 Hz as MOABB does.
Every result is sealed with the exact recording and software version behind it. Change one number and the seal breaks.
Each record stores a SHA-256 hash of the input, the tool version, and an Ed25519 signature over both.
Close your eyes and your brain waves change, plain enough to see on a screen. A tool that misses something that big has no business reading brain signals.
Uses the two one-minute baseline runs in the same PhysioNet set. The expected result is stronger alpha activity, near 10 Hz, over the back of the head with eyes closed.
We scramble the answers and run the test again. A tool worth trusting falls straight to guessing.
Labels are permuted to build a null distribution. How labels are permuted, and how many times, depends on the bench and is not fixed yet.
Every run gets one page, with the score and every reality check it faced side by side. Anyone can open it without special software.
Reports are canonical JSON (RFC 8785), so they can be read without Nasion installed.
The nasion is the dip at the top of your nose, the point EEG electrode positions are measured from. Nasion gives brain signal software the same kind of fixed starting point.
Nasion stands on work other people did first, and we name every piece of it. The control method comes from Görgen and colleagues, the pipelines from MOABB and the recordings from PhysioNet.
Görgen, Hebart, Allefeld, Haynes, NeuroImage 180 (2018) 19-30. MOABB, NeuroTechX. EEG Motor Movement/Imagery Dataset, PhysioNet, ODC-By 1.0.