Ghost bettor_

Every match the model has analysed gets scanned for +EV edges. Anything past the 4% edge floor with a Kelly fraction > 0 gets a paper stake at 0.25× Kelly (capped at 5% of bankroll). No real money is at stake, and none is planned.

This began as a forward-test of whether the model beats the books. That question got answered a faster way — by scoring the model against the closing line directly, on matches already played. The walk-forward floor loses by 10.8σ on football and 13.5σ on tennis; the live model's own measurement is a handful of matches short of publishable and pointing the same way. The full comparison is here. So the log below is not an audition for a betting product. It is the record of what the model would have done, kept public because a strategy that shows only its wins is not a track record.