The 26 GM tenures where the published verdict most disagrees with the record alone. In every case the model is doing the same thing: refusing to let a handful of picks prove anything. Partial pooling pulls thin-evidence tenures toward league average, in both directions.
The record says disaster. The posterior says: five picks is not a body of work.
Nine picks in one 2025 draft, headlined by Travis Hunter at #2. Nothing has had time to happen yet, and at a maturity weight of 0.125 the model declines to read a single class. The widest gap on the board, and almost all of it is the observation window rather than disagreement.
Five picks, one draft, Barkevious Mingo at #6 in 2013. The thinnest tenure in the sample. Now carried at full weight because the window has closed, which makes the model's shrug easier to read: five picks cannot support a strong claim in either direction.
Tim Couch at #1 in 1999 and Courtney Brown at #1 in 2000, at the front of an expansion roster. The record alone is among the worst recorded. The model gives most of it back, reading the situation as much as the selections.
Nine picks in the 2025 draft, Cam Ward at #1. One class, weight 0.125, nothing resolved. The gap is the window, not a verdict.
Bryce Young at #1 in 2023 and Ikem Ekwonu at #6 in 2022. About half the evidence window has elapsed, so roughly half this gap is maturity and half is the model discounting a short record.
Joe Thomas at #3 in 2007 and Braylon Edwards at #3 in 2005. One of the few flagged tenures where the model finishes above the raw record rather than below it, and at full weight that is a genuine upgrade rather than a maturity artifact.
J.J. McCarthy at #10 and Dallas Turner at #17, both 2024. At weight 0.344 most of this tenure is still unresolved and the gap reflects that rather than a judgment.
Six picks in 1980 with nothing above #38. Thin and old: the window is long closed, so what remains is the model declining to draw much from six late selections.
Sam Bradford at #1 in 2010 and Jason Smith at #2 in 2009, premium capital at the top of the draft. The record alone is well below average and the model returns a good deal of it.
Keyshawn Johnson at #1 in 1996 and Kyle Brady at #9 in 1995. The lowest unadjusted record among flagged tenures. Fourteen picks is thin enough that the model keeps a wide interval even at full weight.
Drew Bledsoe at #1 in 1993. The unadjusted record is poor across thirty-five picks and the model recovers part of it.
Marvin Harrison Jr. at #4 in 2024 and Paris Johnson at #6 in 2023. Weight 0.259: this is a tenure the model is explicitly declining to judge yet.
Eric Berry at #5 in 2010 and Tyson Jackson at #3 in 2009. The record alone is below average and the model finishes slightly positive, one of the larger upward corrections among settled tenures.
Eleven picks in the 2025 draft, Ashton Jeanty at #6. A single class at weight 0.125, so the model is holding rather than concluding.
Aaron Curry at #4 in 2009 across thirty-seven picks. The unadjusted record is essentially flat and the model finishes clearly positive, crediting the wider body of selections rather than the headline one.
Zach Wilson at #2 in 2021 and Sauce Gardner at #4 in 2022. Forty picks but only about half the window elapsed, so this gap is still substantially maturity.
Tyree Wilson at #7 in 2023 leading fifteen picks. Short and recent at weight 0.425, and the model treats it accordingly.
Kenny Jackson at #4 in 1984 and Michael Haddix at #8 in 1983. The record alone is poor and the model gives most of it back across thirty-three settled picks.
Chris Long at #2 in 2008 and Adam Carriker at #13 in 2007. Premium capital, a weak unadjusted record, and a model that stays negative while softening it.
The record says genius. The posterior says: one great afternoon is still one afternoon.
Eight picks in 2000, but they included Shaun Ellis at #12, John Abraham at #13 and Chad Pennington. His record alone is among the strongest here. The window closed long ago, so the maturity discount is zero: the entire pullback is the model refusing to credit eight picks that heavily.
Thirty-eight picks across the early 1980s, including Jim Kelly at #14 in 1983. The unadjusted record sits near average and the model lands far below it. Fully settled at weight 1.000, so this is pooling and draft-capital adjustment, not immaturity.
Shawn Springs at #3 and Walter Jones at #6 in the same 1997 draft. The unadjusted record is roughly average across twenty-seven picks and the model is markedly harsher. Settled evidence, so the disagreement is entirely pooling.
Fifteen picks, DeVante Parker at #14 in 2015 and Ja'Wuan James at #19 in 2014. The record alone is slightly positive and the model is not. Fully resolved, so the model is disagreeing rather than waiting.
George Rogers at #1 in 1981 heading thirty-five picks. The record alone is clearly positive and the model settles near zero. Settled evidence, so this is the model discounting early-1980s draft capital.
Jonathan Vilma at #12 in 2004 and Dewayne Robertson at #4 in 2003. The record alone is near average and the model is well below it across thirty-six picks.
Sammy Watkins at #4 in 2014 and EJ Manuel at #16 in 2013. The record alone is a shade above average and the model lands below it. Fully resolved, so this is disagreement rather than patience.