Summary:

We define intelligence as the capacity to acquire and compound resources through accurate prediction. This is the only definition in which each step of self improvement can pay for the next. The recursive loop of self improvement then becomes a probabilistic capital-allocation problem. The model that governs this process is what we call an Economic World Model: a filtration-disciplined forecaster scored on realized outcomes. We support this framework with empirical evidence of a Kaplan based data-scaling law that has an R² = 0.8488 across more than 2 orders of magnitude of data. We also show continued architecture gains from a self-improving auto-research harness, and 16 months of live trading with negligible alpha decay. From these we derive t‑RSI — a confidence-adjusted signal of accelerating self-improvement — currently 1.64.