Throughput, Fitts's law and information rate

Throughput summarizes how much a control interface accomplishes per unit time, and for continuous BCI the natural instrument is Fitts's law: the time to acquire a target of width W at distance D scales with an index of difficulty ID = log2(D/W + 1), and dividing ID by movement time gives a throughput in bits per second. Because it is derived from a speed–accuracy tradeoff over a range of target geometries, Fitts throughput is a robust, task-grounded figure of merit for cursor and reaching BCIs, comparable across systems and against natural movement.

For discrete selection interfaces (spellers, menu BCIs) the common summary is instead information transfer rate in the Wolpaw form, which converts number of classes, per-selection accuracy, and selection rate into bits per minute. It is useful but assumes stationary, equiprobable, independent selections with no error correction, so it can badly misrepresent real usable rate — for instance it credits nothing to an error-correction layer that raises effective accuracy at the cost of speed. Reporting throughput therefore requires stating which definition is used and under what task, since the numbers are not interchangeable.