Model Methodology

Ranking inputs

The engine begins with AP, Coaches Poll, Punt & Rally power, roster talent, schedule strength and returning production. After games begin, weekly snapshots add record, strength of record, yards gained/allowed, points gained/allowed and offensive/defensive efficiency.

In-season ranking weighting

AP is the primary external ranking signal: 45% in Weeks 1–3, 35% in Weeks 4–7 and 25% later in the season. Coaches Poll is deliberately secondary at 10% early and 5% thereafter. Early records are smoothed to avoid overreacting to one result, while every completed game receives an opponent-adjusted performance grade using scoring margin, opponent strength and a three-point venue adjustment. Record, quality wins and on-field performance gain weight as the sample grows. Every weekly model run is stored so ranking movement is auditable.

Hub Model matchup projection

Hub matchup projections are model estimates, not betting odds. They use a predictive team rating rather than AP rank alone. The matchup rating starts with Hub model strength, AP and the Coaches Poll; current on-field performance gains weight through the season. That rating also incorporates scoring margin, yardage margin, strength of record and win/loss performance. A 3.5-point home-field adjustment is applied to non-neutral games.

Performance ramp

The live component is intentionally damped early so one unusual result cannot create a huge line swing. After one game, live performance receives about 8% of the matchup rating; after three games about 30%; after five games about 52%; and after seven or more games about 70%. This makes the projection increasingly performance-driven while preserving a stabilizing prior.

Projection safeguards

Scoring and yardage components are capped before blending to reduce the effect of blowouts and tiny samples. Final projected margins are rounded to the nearest half-point, neutral-site games receive no home-field adjustment, and the margin is converted to a Hub win probability.

Pick ’Em game selection

The production system ranks each week’s slate by ranked-team involvement, model strength, matchup closeness and national relevance, then publishes the 30-game weekly card.

Personal picks

Each user gets one selection per game. Picks are editable until that game’s kickoff, then become immutable and are automatically graded once the result is final.