ASCEND
Engine methodology · Week of August 24 · 8 min read

Inside ASCEND-QX: The AI Sports Prediction Engine, Step by Step

Fixtures are verified against live league schedules, prices are pulled from multiple books and de-vigged, and only selections that clear the edge threshold are published — with the price they were published at.

It starts with a verified fixture, not a projection

Nothing enters the pipeline until the game itself is confirmed. The engine pulls today's schedule from the live league feed, cross-checks it against a second source, and records the fixture identifier, start time and verification flag on every selection it later publishes. If a fixture cannot be verified, it is dropped rather than modelled.

The same discipline governs player props. Team, position and athlete identity come from roster data, so a quarterback prop cannot be priced with running-back inputs. Props are only offered for market types the engine has authoritative inputs for in that league.

Features are standardised before they are weighted

Team and player form, pace, efficiency, rest and travel, injury designations, venue and weather are each converted to a z-score against their own league-season distribution. Standardisation is what allows a two-point line move and a five-percent pressure-rate gap to be compared honestly on the same scale.

The weighted composite is then passed through logistic calibration fitted on out-of-sample periods. The published probability is therefore a calibrated number, not a confidence score dressed up as one.

The market gets a vote — after the vig is removed

Prices are collected across multiple sportsbooks. The engine strips the vig to recover each book's implied probability, then takes a consensus rather than trusting a single number. The gap between the calibrated model probability and that de-vigged consensus is the edge.

Selections that do not clear the edge threshold are discarded, however good the narrative looks. That is why a card can be short. A quota-filling engine is an engine that publishes its worst work.

Staking, then settlement at the released price

Survivors are staked with fractional Kelly, capped well below full Kelly, so a single strong read cannot dominate the card. Each selection publishes with its market, line, price, calibrated probability, edge and stake.

Settlement grades every selection against the price it carried at publication, with pushes reported separately from wins and losses. The public accuracy panels read that settlement ledger directly, so the record on the site is the sum of its rows.

Go deeper

Get the daily card — $12.28 upfront →

More from AI Picks & Insights

Browse the full AI Picks & Insights archive →
Published weekly

The weekly Insights digest

One email a week: the new report, what the model got right and wrong, and the numbers behind it. No touts, no upsells between issues.