Baseline numbers for native (non-proving) inference, measured with the timing
feature on a developer laptop. These are sanity references, not optimized
figures; the proving cost dominates once the zkVM is integrated.
| Model | Inputs | Native inference |
|---|---|---|
| Logistic regression | 4 | < 1 us |
| Decision tree | 3 | < 1 us |
| Tiny MLP (8-8-1) | 8 | a few us |
Proving-time benchmarks will be added once the RISC Zero pipeline lands.