How to read a BCI spec sheet
A short guide to telling validated specs from aspirational ones when you evaluate early-stage neurotechnology, using how MachineDigit labels its own hardware as the worked example.
Early-stage neurotechnology is easy to oversell. A spec sheet can list an impressive channel count, a fast sampling rate, and a long battery life, and none of it has to be measured yet. If you are a developer or researcher deciding whether to test a brain-computer interface, the spec sheet is where you separate what a device does today from what a company hopes it will do. Here is how to read one critically.
Look for a validation status on every line
The single most useful thing an early-stage spec sheet can do is tell you, per line, whether a number is confirmed. A spec with no status is a claim with no evidence behind it. Good disclosure marks each field as validated, in validation, or not yet disclosed, and does not round an unmeasured figure up into a fact. If a spec sheet states everything with equal confidence, treat the whole thing with less.
Separate design decisions from measurements
Some specs are choices the company controls. Connectivity, API design, and supported form factors are decisions, and a team can state them plainly once committed. Other specs are measurements of physical behavior: battery life, weight, latency, signal quality. Measurements have to be tested on real hardware across real conditions. When a measurement is presented as settled before a device has shipped to testers, ask how it was measured and on what.
Watch the channel count and the accuracy claims
Two numbers get inflated more than any others in consumer EEG. The first is channel count, because more channels sound like more capability. In a small, wearable form factor, a low channel count is normal and honest; the trade for something people will actually wear is fewer electrodes, not more. The second is decoding accuracy. Accuracy varies by person, environment, and task, so a single headline accuracy figure with no conditions attached tells you very little.
Ask where the signal is processed
For anything that captures brain-signal data, where the raw signal is processed is both an engineering fact and a privacy fact. On-device processing keeps raw data on the hardware. Cloud processing means raw data is uploaded, which is not wrong, but it changes what the company can promise about your data. A spec sheet that is quiet on this point is leaving out something you should know before you record a session.
How MachineDigit approaches it
On the Wearable page, every spec carries a status. Sensor type and form factors are validated. Channel count and sampling rate are labeled in validation, because they are measured but not yet confirmed across beta conditions. Battery is stated as a target, not a fact. The Platform page says plainly that filtering runs on the device and decoding runs in the platform, so raw session data is uploaded under per-session consent.
None of that is unusual to want. It is just unusual to see stated out loud, and in early-stage neurotechnology, stated out loud is exactly what you should look for.