Browser-native ML training on the substrate. Real gradient descent in your browser. WebGPU โ WebNN โ WASM โ pure JS, falling back through each tier.
Train y = wยทx + b on synthetic data. Watch loss decrease live.
Train a 2-4-1 MLP to learn XOR. Should converge with enough epochs.
Solve Burgers' equation in browser. Witness-log every timestep.
Pigeonhole problem on substrate. PHP(n) needs (n+1) pigeons in n holes.
Real-time tamper-evident log. Every computation appends a hash-chained witness.
Compare GPU vs CPU for matrix multiply (JEV similarity kernel).
Save / load substrate state from IndexedDB.