How The Chairman’s Numbers Actually Work

People assume a pick comes out of pure instinct. Sniff the air, squint at the board, bark once, done. But a hunch isn’t a system, and I didn’t engineer a twelve-season institution to gamble on how the wind smells. Here’s how the machine actually turns football into — hopefully — a smarter number than the one the books printed.

1. The Database — My Cellar

Every serious operator keeps a cellar. Mine is a SQL database that holds twelve seasons of regular-season football — every contest, every score, every drive, every number the books ever posted — pulled together from dependable, free feeds and kept in one place I control. It’s a relational database, not a stack of sticky notes. It keeps the whole history in order so I can pull any single season, any single week, and ask it questions all day.

The point of all that data is one thing: to strip away public bias. Media narratives, home-crowd nostalgia, a color commentator’s hot take — none of it survives contact with thousands of logged boxes. The database doesn’t care who you want to win. It only records what happened, so the model can learn what actually moves a number.

2. The Model — How I Form an Opinion

Once the cellar is stocked, I build a predictive model on top of it. The engine is an advanced gradient-boosted regression architecture (the boring reliable one — simple, low drama, low drawdown). Every game is boiled down to 41 signals — opponent-adjusted efficiency, situational wear-and-tear, what’s been happening on the field and off it — and the model learns which combinations of those signals actually drive a spread.

Then it does something no fan can: it projects its own margin for every matchup and compares that projection to the closing line the market settled on. The difference is the edge — the number of points by which my math disagrees with the books’.

Here’s the discipline that keeps me honest, and I never bend it:

And because I have a system, I have stake discipline too: flat one unit, which is one percent of a bankroll, at the standard line. No doubling up, no “this one feels bigger.” Rhythm and rosters repeat; single results don’t get to rewrite the math.

3. The Week Is a Living Thing — How I Stay Current

Here’s the part most folks miss: the board I set at the start of the week is not the board at kickoff. Between Monday and the whistle, everything moves — the line, the injury report, the weather, the depth chart, the news. A guy who picks once and naps until Sunday is betting a snapshot that stopped being true on Tuesday. So the machine doesn’t make a pick and walk away. It keeps re-reading the week all the way up to kickoff, and so do I.

The single biggest driver of that movement is who’s actually out there. One starter cleared from the report, or one key name shelved overnight, can rewrite a matchup — and the line moves to tell you it did. The same is true along the schedule: cross-country travel, a roof vs. an open sky, a decent afternoon at one venue and wind at another. Nothing about a game is frozen until the final whistle, and then it’s never played again.

And here’s the honesty clause I hold myself to: the closing line is my report card. The market keeps pricing the game as the news breaks, and by kickoff it’s settled on its best final number. I compare what I called at the start of the week to that final number — not to hype, not to the loudest take, but to the number the whole world converged on. Did the competition drift my way, or did I miss something the board saw coming in? That gap between my read and the closing number is where the real feedback lives. It’s not about whether one pick happened to land — it’s about whether, week after week, I end up on the right side of a moving target.

So when you read me kicking a number around all week, that’s not nerves. That’s the discipline: keep the data fresh, keep the variance honest, and let the final closing number settle the argument. A pick worth making is a pick you’re willing to keep updating until the whistle.

4. A Sample Week, Start to Finish

Here’s what it’s actually all my dishes. Take the 2026 slate that just wrapped — preseason, yes, so this is paper only, no stakes, and my model has no regular-season edge in an exhibition. It’s the right snapshot because it shows the way the machine reads an entire board: project a margin, compare to the line, and only speak where the gap clears the threshold.

Read it the way I do. For most of the board, no line posted yet, so no conversation — the column is blank and nobody claims a win in advance. On the games with a number up, the model hands me a projected margin, I keep the closing line, and the difference is the edge.

Two sides of the same discipline: Cleveland at +6.6 looks like a rich mismatch on paper, and Carolina’s +12.3 even richer — but in August, that edge is prep — it’s a depth chart evaluation, half a scripted roster, and starters who mostly won’t play the way they will in September. So it lands on the paper shelf, not the cash register. The same pattern in the regular season is the real thing.

So what would I actually spend?

Real money only lives where the rules are satisfied: regular season, and only when the gap from my model’s margin to the close is at least five points, and a flat unit when it clears. You’ll see that little number in the Shelf column — and when a game is playing for real, the pick carries no “paper” tag. That’s the line between what I study and what I stake.

Preseason, the featured read stays paper — the pick box always says so. When the real season starts, the pick sits on the same engine it always has, and the discipline doesn’t change: project, compare, and only speak where the math is loud enough to deserve your money.

“The nose gets the credit. The database and the model do the arithmetic. I just sign the pick.” — The Chairman