A private research project · practice money only
This journal documents an ongoing experiment. An automated engine forecasts real-world outcomes — daily temperatures, rainfall, index closes — freezes a cryptographic fingerprint of every forecast before the outcome is known, then grades itself against what actually happened. The grades are permanent. The losses stay on the record. Nothing here is offered, sold, or advised.
Every ten minutes the engine repeats one loop: observe → forecast → freeze → grade. Public evidence goes in — a four-model weather ensemble, station observations, exchange prices. A probability comes out and is hashed with its evidence; change one digit of history and the fingerprint breaks. When the market settles, the forecast is graded against reality, win or lose, and the grade can never be edited.
A small practice-money account trades a slice of those forecasts under strict risk caps — not to make money, but to test whether the paper edge survives real order books, fees, and fills. Practice results are practice results; they are reported as such, never as returns.
Each research domain must earn the right to be traded by beating the market's own accuracy over hundreds of graded forecasts. Most fail. That is the system working.
Live grades are momentarily unavailable.