HX 2026.2

How the rating is built

HX is a preseason composite: each feature is z-scored across FBS, then weighted. Georgia opened 2026 at +7.89. A replacement-level Group of Five team sits near 0. Head-to-head uses Elo seeded from that composite.

Weights

Walk-forward Top 25 balanced ≈ 0.646. Only talent and prior rating beat season-to-season noise. QB continuity, coach tenure, and SOS were tested and zeroed — they made the score worse.

  • Recruiting talent1.5

    CFBD team talent composite (247 recruiting strength of the current roster). Real signal — ablation +0.052.

  • Prior-year rating1.5

    Blend of last season’s SP+, Elo, and SRS, each z-scored first. Real signal — ablation +0.037.

  • Recent win-trend1.0

    Weighted win% over the previous four seasons (0.4 / 0.3 / 0.2 / 0.1). Inside the noise band; kept at modest weight.

  • Roster retention0.5

    CFBD returning production (percent PPA + usage). Inside noise. Not the two-deep talent composite.

  • Portal net0.5

    Incoming transfer ratings minus outgoing, this offseason. Inside noise. Roster talent already counts who is actually on the two-deep.

Matchup engine

Each team starts at 1500 + 55 × HX. Home-field is 60 Elo points, dropped on a neutral floor. Win probability is the standard Elo logistic; spread is a quadratic on that same gap. Calibrated on 2019–2025 FBS games at 70.8% straight-up.

Elo = 1500 + 55 × HX
P(home) = 1 / (1 + 10^{−(EloΔ + HFA) / 400})
d       = EloΔ + HFA
spread  = 0.050835 d + 4.5795e-5 d |d|
prev    = d × 0.0534

The old linear map compressed blowouts; this curve was fit on 2019–2023 FBS actual MOV (holdout 2024–2025), not on this week’s Vegas.

Run any pair on the matchup board.

2026 notes

The board is the full 136-team FBS set from the HASHMARK database — recruiting, portal, returning production, SP+/Elo/SRS, and 2026 rosters. Two-deeps are the listed TWO·DEEP charts for 89 programs and a rating-sorted projection for the rest (CFBD publishes no depth chart). Projected two-deeps are labelled as estimates.

Indiana is sixth on the preseason AP ballot after a national title. HX has the Hoosiers eleventh. Talent composite is league-average; the model withholds credit when the roster does not match the résumé.

Georgia and Ohio State are 1–2 here. The AP has Ohio State first and Oregon second; HX flips Oregon and Georgia because Georgia’s talent z-score is the strongest in the pool. Texas Tech’s résumé (prior +1.99) outruns its talent (+0.66), so the Red Raiders sit seventh in HX — above their AP tie at 12th, short of a playoff lock on last year alone.

A Connelly-residual coach-change term is on the board this week (kappa −0.25, with a 15% OC bump). Coach tenure stays zeroed. Week 0 win probabilities are frozen at the pre-adjustment ratings.

Offensive-line mass is on the talent Size board as a measurable. It is not a term in HX. The talent composite is the depth-weighted 247 of the two-deep, transfers included.