Edge without staking discipline is noise. This engine sizes every Ascend play against your actual roll using fractional Kelly, de-rates for correlation and calibration drift, models the drawdown distribution, and tells you to shrink before variance does it for you.
The schedule this engine runs on. Every run writes a heartbeat, so if a cycle is late you can see it on the status board rather than guessing.
7 weighted inputs, published in full. Weights sum to 1.0 and are refit against out-of-sample periods — never against the window being reported.
The realised edge of each tier over its trailing 250 plays becomes the base input — stated confidence is not.
The gap between stated and settled probability by bucket de-rates any tier that has been over-claiming.
Shared game, player and market exposure is collapsed so six plays on one game are not sized as six plays.
Return dispersion at the prices actually available, plus the stake that clears without moving the number.
The account is placed inside its modelled drawdown envelope, and exposure falls as it deepens.
Fractional Kelly with a hard cap produces the day's stakes, published with the reasoning behind each one.
What members ask before unlocking AI Bankroll Manager.
A sized card: a stake per play as a fraction of your roll, correlation caps across same-game exposure, and where the account currently sits inside its modelled drawdown envelope.
The engine uses fractional Kelly with a hard cap, and it de-rates for ticket correlation and calibration drift. Full Kelly is optimal only if your probabilities are exactly right — they never are.
It sizes from settled results: the realised edge of each tier over its trailing plays and the measured gap between stated and settled probability, so a tier that over-claims shrinks automatically.
No. It keeps losing stretches survivable by bounding exposure. A thin genuine edge still produces losing weeks, and no staking rule changes that.
Ascend publishes quantitative research, not investment or betting advice. Every published call can lose, past model performance does not predict future results, and sizing is your decision. Never risk capital you cannot afford to lose.