A measurement-efficient runtime that replaces the optimizer in your loop. When measurements are scarce or your device drifts, it reaches a usable answer in a fraction of the budget — watch it settle ahead of SPSA, live, below.
The runtime replaces gradient estimation with a residual-feedback update. That edge shows up in three places — and the client tells you honestly when you’re outside them.
Below the budget crossover it leads SPSA, parameter-shift, COBYLA and Adam on anytime quality.
Where gradient estimators can’t even run, it tracks a drifting target at the noise floor.
Its momentum acts as a predictive term, holding lock under control delay that breaks PI and PID.
Get a key, drop the client into your loop in one line, and run it on your own workload. The method runs on our endpoint — nothing lands on your disk.
Get your key →