How the model works
A transparent impact model
Compute for Good turns four open drivers into a single impact density, then applies diminishing returns so the optimizer spreads scarce compute instead of dumping it all in one place. Illustrative only.
Driver weights
| reach | 0.35 |
| urgency | 0.25 |
| tractability | 0.25 |
| evidence | 0.15 |
Weights sum to 1. Marginal impact of the h-th GPU-hour = density × e^(−h / saturation).