AlphaHand
Featured run
2-m16 · 2026-03-19
89.79% action · 87.01% finger on non-REST
Primary holdout
84.66% joint accuracy
98.37% REST TPR
18.57% applicability FP on true REST
Pseudo-live replay
93.32% would-send precision
86.64% committed joint accuracy
0.12% false REST actuation
Published corpus
4,953 test windows
4,532 non-REST across 2 published runs
Results
Current metrics, figures, and replay summaries. Evaluation contexts are labeled separately.
Deep dive
The longer model-selection story: sweep scale, ablations, replay gates, and the dated trail behind the featured run.
How it works
Pipeline, preprocessing, windowing, and model notes live there instead of being repeated across the site.
Paper
Manuscript status page and placeholder PDF while the actual pre-print remains private.
Featured figures
The current featured bundle includes confusion, calibration, replay, and latent-space artifacts. Start on the results page for the full context.
PCA · Full dataset
Each point is one EEG window, projected into three principal components and colored by the labeled finger.
Separated regions suggest structured learned organization, while overlap marks similar or harder windows.
Latest update
See the updates page for the dated release note.