Consumer neurotech and wellness BCI/EEG

Scalp EEG is a real measurement of a real thing, and most consumer disappointment comes from misreading which thing. The signal is not neural firing. Action potentials are too brief and too randomly oriented to sum. What reaches the scalp is the slow extracellular field of synchronised postsynaptic potentials in cortical pyramidal neurons, which are aligned perpendicular to the cortical surface and therefore add coherently. Roughly a square centimetre of cortex must be active in step before anything is detectable, which is why EEG sees rhythms and population states rather than computations.

Why the skull is the hard part

Between that source and the electrode sit cerebrospinal fluid, skull and scalp. Bone has low conductivity relative to the tissue around it, so the field is both attenuated — arriving at 10–100 µV, orders of magnitude below the mains hum in an ordinary room — and spatially smeared. Volume conduction means each electrode reads a weighted mixture of many sources, and recovering the sources from the mixture is an ill-posed inverse problem. It becomes tractable only with many electrodes; it is not solvable at all with two or four. A consumer headset cannot tell you which brain region did something, and no software can retrofit that.

Dry electrodes and the impedance penalty

Clinical EEG uses conductive gel to bring the electrode-skin contact impedance to a few kilohms. A dry electrode skips the gel — the whole point of a wearable — and typically sits one to two orders of magnitude higher. Impedance itself is less damaging than impedance mismatch between electrodes, because differential amplification rejects shared interference only when both inputs see similar source impedance. Mismatch converts common-mode noise into apparent signal. Dry contacts also drift: the electrode-skin half-cell potential changes with sweat, temperature and pressure, producing slow baseline movement in exactly the low-frequency band where delta activity lives.

What a frontal montage actually sees

Most consumer devices sit on the forehead, the one accessible spot without hair. It is also the worst place for interference. Eye movements and blinks generate corneoretinal dipole potentials of hundreds of microvolts — larger than the EEG. The frontalis and temporalis muscles emit broadband EMG that overlaps the beta and gamma ranges, so a “gamma focus score” from a forehead sensor is frequently jaw and brow tension. Blind source separation helps, but decomposition needs at least as many channels as sources; on two channels there is nothing to separate.

Within those limits, some readings are solid. Occipital alpha at 8–12 Hz appearing on eye closure is among the most reproducible findings in electrophysiology. Frontal slow-wave activity is genuinely frontally maximal, so single-channel sleep staging can separate wake, NREM and REM with moderate agreement against full polysomnography, which the AASM defines as EEG plus EOG plus EMG.

The gap is between measurement and intervention. Neurofeedback is hard to blind, and sham-controlled trials often fail to show effects specific to the trained parameter. That is why these devices are marketed under general-wellness framing rather than as medical claims — a regulatory position, and also an accurate description of the evidence.