We Said “Probably Wrong” on Friday. Notre Dame Lost on Saturday.

The question we ask every Sunday: what did we claim this week, and did reality agree?

The scoreboard. Six issues, two season-long picks to account for: the Bills to win the Super Bowl at roughly 15%, Notre Dame to win the CFB title at roughly 16%. Both remain technically pending. But Notre Dame lost to BYU 28–24 on Saturday night, one day after Issue #202 published those numbers. Say it plainly: we called them champions. They lost the next day.

Here is the important part, and it is not an apology. Our Issue #202 title was “Book It. We Are Probably Wrong.” The 16% was printed in the second paragraph. A 16% call does not predict an undefeated season — it predicts exactly what happened: a team that very likely loses somewhere, whose exact chance of winning it all was one in six. Notre Dame is now 6–1. They can still win the ACC championship game and earn a CFP bid. The pick was not wrong because it came with its own uncertainty attached. But we owe you the update, and here it is: their title probability is not 16% anymore. It is considerably lower.

Why we bother: a prediction nobody grades is just a story.

The Week in Review: Issues #198–203

1. #198Oct 12 — Twenty-Four Times in a Hundred. This Was One of Them. — Wisconsin beat Penn State as a 24% underdog. The number was not wrong. This is what a correctly-priced 24% shot looks like: it comes in roughly once every four attempts, and you cannot tell beforehand which attempt it is.
2. #199Oct 13 — We Said Ninety Wins. They Won Seventy-Four. — Our five Mets forecasts for 2026, scored against the final 74–88. The gloomiest number was the most accurate. The one with a credible interval attached was the only forecast that contained the answer. The unadorned number was nearly right; the reasoning laid on top of it was the error.
3. #200Oct 14 — He Inherited the Roster. He Wins One Game Fewer Than It Should. — A new coach’s win total correlates with the talent he inherited, not the talent he built. By year five the relationship inverts. Until then, you are measuring the school more than the coach.
4. #201Oct 15 — A Coin That Hits Like Ohtani Slumps More Than He Does. — Ohtani went 0-for-21 in the NLCS. A coin weighted at his career average would go hitless in 21 consecutive at-bats about 1 time in 2,300. That is rare enough to be worth checking — and not rare enough to require an explanation beyond chance.
5. #202Oct 16 — The Bills and Notre Dame Will Be Champions. Book It. We Are Probably Wrong. — The Bills are the NFL’s likeliest champion at about 15%. Notre Dame is college football’s second likeliest at about 16%. We published both picks knowing an 84% chance meant we were probably wrong. Notre Dame proved it the next night.
6. #203Oct 17 — The Upstarts Won 103 Games. — Milwaukee swept Los Angeles 4–0. The Brewers had the best record in baseball. The sweep was a 1-in-14 shot; the Brewers winning the series was a 54% proposition. When someone calls this an upset, ask which part they mean.

The Prediction Scorecard

Issues #198–201 and #203 contained no forward forecasts — five analytical and retrospective pieces. Issue #202’s Bills and Notre Dame picks remain the active claims. Notre Dame’s title probability is lower than 16% after Saturday’s loss; we will recompute and report when the CFP picture clarifies.

Five of the Top Seven. The First Time Ever.

Five of the top seven teams in the AP Poll lost on the same Saturday. It is the first time that has happened. Here is the list: No. 1 Georgia fell to unranked Auburn 27–24 in Athens. No. 2 Notre Dame fell to No. 8 BYU 28–24 in Las Vegas. No. 6 Alabama fell to No. 14 Tennessee 26–10. No. 7 Texas fell to unranked Florida 27–24 in Austin. And No. 5 Indiana lost to No. 4 Ohio State 19–15, which is not an upset but does close the door on Indiana’s CFP momentum.

The broadcast is calling yesterday historic. That part is correct. But “historic” is a description of the record books, not of how improbable it was. Those two things are not the same.

Here is a rough calculation. Assign each top-seven team a realistic win probability against their Saturday opponent, based on the pregame spread. Georgia was about a 10-point favorite at home against Auburn — that implies roughly a 74% win probability. Notre Dame was about a 3.5-point underdog at BYU (away, at night), which flips to a 59% win probability for the Irish. Alabama was about a 6-point favorite against Tennessee — roughly 67%. Texas was about a 9-point favorite at home against Florida — roughly 73%.

So on a random Saturday this was about a 1-in-100 event. There are roughly 13 Saturdays in a college football season. Over ten seasons — 130 Saturdays — the probability that this specific combination of losses never occurred is about 27%. Which means a 73% chance it would happen at least once across ten years. We got the “at least once.” It took until 2026.

The lesson is the same one Issue #198 ran on Thursday, using Wisconsin and Penn State: a low-probability event is not an impossible one. You cannot look at five upsets and conclude the forecasts were wrong. You can look at five upsets and conclude that five individually-unlikely things happened at once, which is the expected outcome of running 130 Saturdays in a row.

What Readers Read · October 10 – October 16

Counts are unique pageviews per GoatCounter, with author visits excluded. Pulled 2026-10-17.

What to Watch Today — and Why Those Games

The NFL card today is ranked by nearness to a coin flip. The spread is the market’s best estimate of the gap between these clubs. A game at pick’em or minus-1.5 is two teams the market genuinely cannot separate. Those are the games where a single missed assignment, a lucky bounce, or a third-down conversion holds the most information about which club is actually better.

We hold no NFL model. Six weeks is not a model. What you see below is the published spread as of this morning; a 50-50 game is labeled for what it is. The Bills’ game is included at the bottom because the pick is ours and you deserve to know where it stands.

Spreads from market sources as of Sunday morning. Win probabilities use a 13.5-point NFL standard deviation: a 1.5-point favorite wins about 54% of the time, a 3-point favorite about 59%, a 5.5-point favorite about 66%. Where we hold no model of our own, we say so rather than dressing a spread up as a forecast.

The Postseason: One Series Done, One Tied

Milwaukee swept Los Angeles in the NLCS and advances to the World Series for the first time since 1982. Issue #203, published yesterday, has the full analysis. The short version: the Brewers won 103 regular-season games, scored 4.8 runs a game, and held the Dodgers to 11 runs in four games against a lineup that averaged nearly 5. The sweep was a 1-in-14 shot; the Brewers winning the series was the most likely outcome.

In the American League, Cleveland and Tampa Bay are tied 2–2. The Rays evened the series with a win Saturday. Game 5 is tomorrow in Cleveland. The Guardians went 85–77 in the regular season — the lowest win total of the four remaining clubs when Milwaukee’s NLDS run began. Of the four teams that started this postseason, Cleveland is the one the scoring chart from Issue #203 described as “flat.” Their October run-scoring rate almost exactly matches their September rate. They are not losing their bats the way everyone else is. That is either a good sign or a small sample.

Over-Reactions and Under-Reactions

What to Take Home, and the Road Ahead

The week taught three things worth holding onto. First: the Mets scorecard in Issue #199 showed that the gloomiest number was the most accurate, and that every argument against it — healthier arms, an easier schedule, regression to the mean — cost us wins. Reasoning against your own model’s output requires new measurements, not new reasons.

Second: calibration is not about one game. Issue #198 walked through why Wisconsin beating Penn State at 24% odds is not evidence that the forecast was wrong. The same logic applies to everything that happened Saturday. Each of those losses was unlikely. None of them were impossible. The way to test a forecast is across many predictions in a similar probability band — not from a single Saturday.

Third: the Notre Dame pick in Issue #202 shows what an honest probability statement looks like in practice. We named them champions. We printed the exact odds against it in the same paragraph. They lost. The statement was still correct, and updating the probability downward is what the model is supposed to do. That is not a failure of forecasting. It is forecasting working as intended.

The queue has one piece ready: a deep look at schedule strength and how to tell whether a team has earned their record or inherited it from a weak schedule. The payroll explosion piece is behind it. Both are durable enough to wait for the right editorial slot.