LINX scores a running back's skill using yards after contact, broken tackle rate, and O-line-adjusted yards before contact, measured against a 100-carry-minimum sample from 2018 to 2025. It's built to strip out how good their blockers are, so what's left is closer to the runner's own talent. The fantasy engine then adds volume, role, and age on top of that skill score to project PPR points. Weights were fit on 2018–2022 seasons only, then scored on 2024 and 2025 outcomes the model never saw.

Running back year-to-year prediction has a low ceiling for everyone, us included. Our honest blind number is r = 0.29, and the improvement over the simpler version of LINX is suggestive rather than statistically significant — the confidence interval on that gain crosses zero. We'd rather say that than round it up.

WRX scores a receiver on three things he controls: catch-rate-over-expected (how often he wins the ball, adjusted for how far downfield it's thrown), broken tackles per reception, and drop rate. Weighted .70 / .20 / .10 in that order.

Separation is deliberately not in it. We pulled real Next Gen Stats separation and tested it. It is the most stable receiver skill we measured — year over year r = 0.60, better than any input we kept — and it predicts next season's fantasy points essentially not at all (r = −0.02, p = 0.70, and +0.001 R² on top of what we already had). Getting open is a real skill that doesn't convert into fantasy points, because points come from volume and touchdowns, which come from role. So we dropped it, along with yards-after-catch-over-expected, which carried no signal either. The same result held independently at tight end.

QBX scores a quarterback on rushing production, passer rating, and a per-game production rate (pass yards and touchdowns), weighted .47 / .28 / .25, each turned into a z-score by season and blended with a recency weighting across 2018–2025 (roughly 60% 2025, 30% 2024, 10% 2023 for a player with a full sample).

We started with six inputs and let the data cut three. Re-fitting the weights from scratch at four different cutoff points, accuracy, arm strength, and EPA per dropback came out at exactly zero every single time — EPA because it's redundant with passer rating, the other two because they don't translate into fantasy points. Rushing carried the most weight at every cutoff. For quarterback fantasy value, running matters more than any passing trait, and it's also the most repeatable thing a quarterback does.

The fantasy projection is a separate, simpler layer on top of that skill score: recency-weighted per-game rate stats (pass yards, pass TDs, INTs, rush yards, rush TDs) multiplied by a researched 2026 games-available figure for that player's role. QB scoring is the same across PPR, Half, and Standard, so QB rows don't move when you flip formats.

The weights above were validated out of sample — refit at four separate cutoffs and scored on the season after each — but QBX as a whole hasn't been through the same strict single train/test holdout as LINX and WRX, so read it as a skill snapshot rather than a validated predictor on its own. Most of the real uncertainty in the QB projection lives in the researched 2026 role and games-available assumption, not in QBX itself.

TEX scores a tight end's per-play receiving skill from separation, yards after the catch above expectation, broken-tackle rate, drop rate, and catch percentage — the same theory-weighted approach as LINX and WRX. But leave-one-season-out testing found TEX adds essentially no accuracy on top of a simpler opportunity model: last year's per-game production, projected 2026 target volume, and a touchdown rate regressed toward the league TE average on a 45-target prior. So the TE board is ranked by that opportunity-driven projection. TEX is still shown next to it as context on each player's skill, it just isn't what decides the rank.