RK logoReed KnapperCOLLEGE FOOTBALL · RKR MODEL
College football analyst · host · model builder

Reed
Knapper

I cover college football with numbers behind every opinion. Every week I break down the games, rank the teams and make my picks, all backed by Reed Knapper Ratings, a predictive model I built from the ground up to rate all 138 FBS teams and project every game.

Live model · updated hourly on game weeks
Teams rated
138
Every FBS team, re-rated after every game
Backtest, all rules
60.7%
347-225 at the opener, 2021–26
Data behind it
2014–
Seasons of game, line and play-by-play data
Live record
—
Every pick graded in public
About me

Who I am

I'm Reed Knapper, a college football content creator. I host a weekly college football show where we walk through the College Football Playoff picture, preview the biggest games of the week and argue about who's for real. I also run a college football YouTube channel and social pages, where the picks, power ratings and upset alerts go out every week.

My NITT PICKS episodes go deep on Penn State, and the blue and white all over this site is a nod to Happy Valley.

I'm also part of the CFB Redzone Crew, a group of college football creators who put our best bets side by side every week and keep score.

What ties it all together is the model. Instead of picking games on gut feel, I built a system that projects every game, tracks itself honestly and gets better every week. It's the foundation of everything I post.

The goal

What I'm building in college football media

01

Analysis you can check

Every take should have a reason you can see. When I say a team is underrated or a line is off, there's a number behind it, and the number is on this site for anyone to look at.

02

Every pick graded in public

No deleting the misses. Every official pick is logged before kickoff and graded afterward, win or lose, along with whether we beat the closing line.

03

Analytics for every fan

Success rate, explosiveness, line movement: I want to make the stuff the sharps use simple enough for any fan to enjoy on a Saturday, through graphics, the show and this site.

Reed Knapper Ratings (RKR)

How I built the model

RKR started as a spreadsheet and became a full model that refits itself every week. Here's how it turns raw data into a projected line for every game.

Collect the data

Games, betting lines, play-by-play efficiency, Elo, roster talent and returning production from CollegeFootballData.com, going back to 2014. Every backtest uses only what was known before kickoff.

Rate every team

Each team gets opponent-adjusted offense and defense ratings for success rate, explosiveness, rushing and passing efficiency, line yards, stuff rate and short-yardage success, updated after every game, plus a margin-of-victory rating and Elo.

Learn from the market

A market rating built from past closing spreads teaches the model what Vegas already knew. A travel factor, faster in-season updates and a preseason starting point from talent round it out.

Fit the weights

A ridge regression on a decade of games decides how much each piece matters, refit season by season so the model is always tested on games it never saw. In the portal era, recent seasons count more.

Turn edges into picks

A pick only counts when the model's line beats the opening line by enough: 4+ points in power-conference games, 4–6 in Group of Five games, and unders 5+ below the total. No official picks in Weeks 1–3 or bowls, and a big line move against a pick drops it to a lean.

Re-check every hour

Lines refresh hourly from 10 sportsbooks until kickoff, and a live in-game model built from 6,400+ games drive by drive updates during games. Each Sunday the ratings refit and next week's projections post.

The build

From spreadsheet to live model

The spreadsheet

The first version lived in a Google Sheet. It produced picks, but an honest audit showed the best bets were basically a coin flip, and bigger edges weren't winning more often.

Rebuilt on a decade of history

I started over with a historical dataset going back to 2014 and a walk-forward backtest, so every rule had to prove itself on seasons the model hadn't seen.

Rules that survived testing

Only bet at the opener, shop the best line, unders only on totals, and treat huge Group of Five edges as model error. Ideas that didn't hold up in both test periods were thrown out.

v3.1: learning from the market

Adding a market rating and travel made the model's lines sharper. Live lines, props, upset alerts and an in-game model were built on top of it.

v3.4: built for the portal era

Rosters flip way faster now. Teams went from about 6 incoming transfers each in 2021 to 22 in 2026, and returning production keeps dropping. So I re-tested the model on 2022–26 only. What mattered: giving recent seasons and recent games more weight, and trusting returning production less early in the year. What didn't: portal rankings, transfer QBs and first-year coaches added nothing the model wasn't already seeing. In that test v3.4 went 231-145 at the opener, up from 219-161, with smaller misses in 4 of 5 seasons. That's the version running today.

Track record

The numbers, honestly

What the testing says

  • 347-225 (60.7%) in the 2021–26 backtest at the opening line, across spreads and unders.
  • When a pick qualified at the opener, the market moved toward the model 58% of the time.
  • On seasons the rules weren't tuned on (2017–19, against the closing line), it went 144-138 (51.1%). That's the realistic floor, and why I track closing-line value every week.

This season, live

  • Live record: loading…
  • Beat the closing line: loading…
  • Every pick, win or lose, is on the record page.

The goal is to beat the market over the long run. One good or bad week doesn't prove anything either way, so it's all out in the open.

Explore

Inside the site