← CFB Favorites · After The Timeout

CFB Best Ball · Favorites

J'Koby Williams, RB Texas Tech

He is in three of my five teams. My own board has him 253rd. This is the one where I am arguing with my model, so I went and found out which of us is wrong.

60% exposure across 5 drafts · ADP 146.7 · my board has him 253rd

Update, Sept. 17: written before the season from my first five drafts. Across all 16 drafts he's on 4 (25%), ADP 154.6, my board 240th. The test I set below, 4 to 5 touches a game, has already been passed: 21 and 22 touches in his first two games (ESPN box scores).

Every other piece in this series is me finding a player I think is priced too low. This one is the opposite, and it is more useful.

J'Koby Williams is on three of my five teams. My board has him 253rd, which is a hundred and nine spots below where the market takes him. I built that board. I keep drafting the guy anyway.

So either my hands are wrong or my model is. Worth finding out which.

Part 1: What I paid

Sixty percent exposure. Same as Hasley, and Hasley is a player my board loves.

His ADP is 146.7, round twelve or thirteen depending on where the room drifts. My board says 253rd, which in a twenty-round draft means it does not want him at all.

That is the widest disagreement between me and my own rankings anywhere on my teams.

Part 2: Why the model hates him

I went and read my own code, which is a strange sentence to write.

My CFB board assigns each player a role inside his team's depth chart, then weights his projection by it. Role one is full value, role two is 72 percent, role three is half. A running back has to clear a replacement level of 167.2 points to be worth a pick at all, because superflex rosters are deep and running backs are plentiful.

Williams projects 134.8. The board calls him the RB2, multiplies to 97.1, and subtracts replacement. Minus seventy. Dead last tier.

Here is the part that matters. The model decides he is the RB2 by looking at ADP. Cameron Dickey goes at 65.5, Williams at 146.7, Quinten Joyner at 214.1, so the board sorts them one, two, three and applies the penalty.

That is circular. My board read the market's opinion of his role, applied a discount for it, and then reported the result back to me as if it had found something. It has not evaluated him at all. It has laundered ADP through a multiplier.

I do not get to call that a finding.

Part 3: What the model cannot see

Williams caught 35 passes last season for 388 yards. Dickey, the back who carried it enough to run for 1,124 yards and 14 touchdowns, caught 24.

The RB2 out-caught the thousand-yard starter by eleven receptions.

That matters more in this format than it looks. Underdog is half PPR. A catch is half a point before the ball has gone anywhere, and receptions arrive more reliably week to week than carries do, because they do not depend on the game staying close. A back who catches the ball is a back with a floor in weeks his team is behind.

My role weight is a single number applied to the whole projection. It cannot tell the difference between a runner who is second in line for carries and a receiver who happens to line up in the backfield. Williams is closer to the second thing, and 0.72 of a blended projection does not describe him.

He also averaged 5.6 a carry for 868 yards while being the guy who was not getting the ball, and took a kick 99 yards for a score.

Part 4: The case against me

If I only wrote the part above, this would be advocacy rather than analysis.

Quinten Joyner is back. The USC transfer tore his ACL in a preseason scrimmage last August and missed the entire season. Joey McGuire had him cleared for camp on June 1. A healthy Joyner makes this a genuine three-way committee, and three-way committees are where best ball running backs go to die.

Dickey is not going anywhere. He is on the Maxwell Award watch list. He led the Big 12 in total touchdowns. Nothing about last season suggests Texas Tech is looking for a reason to take carries off him.

And the depth is the point. The most-quoted line about this backfield is that Williams would start for the vast majority of teams in the country and might be third string at Texas Tech. That is a compliment and a warning in the same sentence.

So my model's conclusion, be careful here, may well be right. Its stated reason is still wrong. Those are different failures and it is worth keeping them apart.

Part 5: Why I clicked anyway

Round thirteen is where I am buying outcomes, not floors.

At that price I am not paying for the median. I am paying for the branch where Joyner's knee is not all the way back, or where Dickey misses two games, or where Texas Tech simply keeps doing what it did last year and lets the passing-game back stay on the field. Williams was already producing 1,256 scrimmage yards as the second option. The version of him that gets even a third of the carries is a weekly starter in this format.

That is the shape of a best ball pick in the double-digit rounds. I am not trying to be right about the average outcome. I am trying to own the tails cheaply.

Sixty percent is probably too high, though. Three of five is a real bet and I made it in a room where I was also taking Dickey in one draft, which means one of my teams owns both halves of the same committee. That is not a hedge. That is paying twice for one backfield.

Part 6: What would change my mind

If Joyner looks like himself in camp, this pick gets materially worse and I should be at one of five, not three.

If the first two games come and Williams is at four or five touches, the receiving role I am paying for did not survive Joyner's return, and there is nothing to wait for.

And if reporting out of Lubbock this month describes a clear one-two with Joyner second, I am the one who is wrong, not the board. It would still be wrong for the reason I gave in Part 2, and I am going to fix that regardless of what happens to this pick.

Part 7: The fix

I am changing the board, not the pick.

Deriving role from ADP has to go. It makes the model a mirror. The replacement it should be measured against is a receiving back's, not a runner's, and the weight should respond to how a player is actually used rather than to where he is drafted.

Until that is done, treat every role-two running back on my CFB board as unranked rather than as ranked low. The number in that column is telling you what the market thinks, in my handwriting.

Get the next spotlight

Leave your email and we will tell you when a new college best ball write-up or tool goes up. No set schedule, no spam, unsubscribe whenever.

Website

Send it

Exposure figures from my own Underdog draft exports. ADP from Underdog best ball rankings. Statistics and quotes from published reporting. Nothing here is betting advice, and I am telling you what I did, not what you should do.