Papers, infographics and notes on how college football works, built from the same data as the rankings.
A fixed combination of Lean C1, observed quarterback-and-line roles and robust scoreboard strength. Enhanced historical comparisons favor its error counts, while uncertainty, source vintages and adaptive development require a separate prospective test.
A 247-input, market-free rating with close-strength weighting, annual earlier-only selection and explicit fallbacks. Historical error counts improve on the reconstructed AVR control, but uncertainty does not establish an accuracy gain or all-four incumbent superiority.
AVR’s Division I rating and tiers, evaluated against FPI, Massey MAS, Sagarin Rating and SP+. Matched historical ordering favors AVR over MAS; all-four parity is unproven, and margin results favor FPI and Sagarin.
Following the Talent · Research summary
Comparison periods: 2016–2019 and 2022–2025. Value weights use performance from the season before a move.
1.9% → 23.5%
Annual share of returning players’ proven value that changed programs.
Denominator: prior-season proven value among players observed in consecutive seasons.
77%
Share of transferred value moving to a higher-ranked program.
95% confidence interval: 75–80% in 2022–2025. Higher rank is measured on the previous season’s AVR board.
42.9% → 43.1%
Top-50 share of measured FBS talent, including estimated contributions from unproven players.
Crediting transfers to origin programs reduces the post-period share by 1.7 percentage points (95% CI 0.5–2.8). This is an accounting comparison.
14.4%
Subsequent draft rate among G5 regulars who moved to Power programs. Selection into transfers limits causal interpretation.
2022–2024 moves. Difference: 6.4 percentage points (95% CI 2.9–9.9). Matching uses origin category, prior grade and unit.
AVR’s v1.1 tiers use fixed rating bands in points above an average Division I team. Rank measures ordering; a tier identifies a strength band. The flow estimates in this paper use rank bands and conference categories, rather than these rating tiers.
Boundaries include the lower endpoint and exclude the upper endpoint. Tier-based transfer estimates require classifying both programs on the same prior-season board; the 77% rank statistic does not measure tier advancement.
Descriptive and matched observational estimates. Effects of NIL and transfer reform are not separately identified. Four seasons per comparison period limit season-level inference.
Andy Volpe · Working paper · PFF weekly grades, CFBD records and AVR ratings. The full paper also reports partial 2026 observations; those are excluded from this summary.
Performance-weighted player mobility increased after 2021, while the top-50 share of measured talent remained similar. An observational study of talent flows, team strength and draft outcomes.