Every prediction.
Measured over time.
Learn which sources help. Freeze the methods. Check what actually arrives. Build a searchable history of performance, disagreement and change.
AUTOMATIC FIELD-ACTIVITY PILOTMatched history
Fresh predictions
Outcomes checked
The question we are testing
The first audit selects the largest matched field-activity cohort using non-overlapping windows, without selecting for the best result.
No protocol frozen yet.
Fresh-data comparison
Brier score measures probability error. Lower is better. All six methods use the same cases and stay frozen throughout this pilot.
Experimental results require review. This pilot does not change production weights automatically.
Historical replay
An earlier time window fits the methods; a later window checks them. Only outcomes available at the cutoff enter training. This is descriptive history, not proof of future performance.
Evidence you can follow
Training cases, frozen source profiles, experimental estimates and outcome reports are saved privately into the OMNIA dataset with source links. Original reports remain available when a later correction invalidates a result.
This pilot measures recorded activity counts. Both models share upstream observations. It does not establish real-world event truth, physical EFVT measurements, investment performance or universal credibility.