Oklahoma Had a One-in-Three Shot. It Hit.

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

The scoreboard. Six issues, zero gradeable predictions: nothing hit, nothing missed, nothing pending. All six pieces this week were analysis, retrospective, or answers to reader questions. That is worth saying plainly rather than quietly skipping over. A week without forecasts is a legitimate kind of week.

There is nothing to account for here. No miss to name. No hit to celebrate. If that bothers you, the right place to put that energy is next week, when the queue returns to analytical pieces with forward claims. A prediction nobody grades is just a story — but not grading something you never predicted is not a dodge.

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

The Week in Review: Issues #191–196

1. #191Oct 5 — Three Hundred Million Tickets. One Expected Winner. — There are 292,201,338 ways to fill in a Powerball ticket. If every American bought one, you would expect a single winner. This is not a piece about foolish buyers. It is about what the house calls the money it keeps.
2. #192Oct 6 — Defense Moves More Than Offense. Measure It Better and It Does Not. — Counted in points, defense is 4 percentage points likelier to jump a full standard deviation, and the interval clears zero. Count the same clubs in yards per play and the gap falls to 1.5 and stops being a gap at all.
3. #193Oct 7 — The New Coach Recruited None of Them. We Start Counting Anyway. — In year one of a new regime, 0% of the roster came from the new staff. Half his by year three. All his by year five. And the research says firing a middling coach makes the team worse.
4. #194Oct 8 — He Threw 104. The Box Score Said He Climbed Three Levels. It Did Not. — A 22-year-old in Las Vegas threw seventeen fastballs, none under 101.5. His season line reads Single-A to Triple-A. The transaction log says something else, and the difference is the whole denominator lesson.
5. #195Oct 9 — Ninety-One Chances. Six Upsets. — Unranked teams have beaten ranked opponents only six times in five weeks, the fewest since 2019. The six is right. What nobody printed: there were more chances at an upset this year than in any since 2019, and the favorites held anyway. The probability math on what a “normal” upset count would be.
6. #196Oct 10 — Kickers Make 80% From 43 Yards. That Told Chicago Nothing. — Gene, a Bears fan, asks the right question. If the long run is so reliable, why does it feel like it quits at the last second? It never quit. It was never a fact about one kick.

The Prediction Scorecard

Issues #191–196 made no forward forecasts. That is five methods and analysis pieces plus one reader-question response. The queue ahead has pieces with quantitative claims; the scorecard will have something to grade next Sunday.

The Improbable Thing That Happened Yesterday

Oklahoma beat No. 1 Texas 27–24 in Dallas on a walk-off field goal. The Sooners entered as a 6.5-to-8.5 point underdog, depending on which line you read. The market-implied win probability for Texas ran between 65% and 70%. Oklahoma’s chance was somewhere between 30% and 35%.

The broadcast is calling it a shock. It is not a shock. A 35% event happens roughly one time in three. If you walked up to a coin that came up heads 65% of the time and it came up tails, you would not say the coin is broken. You would say: yes, that happens.

Here is the number that actually matters. There were roughly 90 rated college football games on the card yesterday. Add up every underdog’s chance of winning its own game and you get the number of upsets a normal Saturday produces. Yesterday’s card carried three top-ten teams who lost — Texas, Oregon (to UCLA, as underdogs around 25-30%), and LSU (to Kentucky in overtime, similarly priced). None of that is improbable if you think about the card as a whole instead of one game at a time.

The surprise was never that a top-ten team lost. The surprise is only ever which one.

The arithmetic I still cannot give you. Each Sunday this section is supposed to sum every underdog’s pre-game win probability, from our frozen ratings, and compare it against the actual upset count. That requires a snapshot taken before kickoff on Thursday; the Week 6 snapshot was not taken. Printing the arithmetic on ratings that have already absorbed Saturday’s results would be how you convince yourself you saw everything coming. I will not do that. The lesson holds even without the math. But the math would have shown you how ordinary yesterday was — and that is the part worth seeing. We are working to fix the snapshot cadence before next week.

Other results worth noting. No. 2 Georgia held off No. 6 Alabama 20–17. That is not an upset — the second-ranked team beat the sixth-ranked team. Indiana beat Nebraska 20–17. Notre Dame continues its winning streak; the Irish are now 6–0. The CFB Playoff picture tightens at the top every week that Georgia and Notre Dame both win, because they are on a collision course.

What to Watch Today — and Why Those Games

The NFL card today ranks by nearness to a coin flip, not by marquee matchup. The spread is the market’s best estimate of the gap between these clubs. A game at pick’em or −1.5 is two clubs the market genuinely cannot separate. Those are the games where a single play — a missed assignment, a lucky bounce, a Gene-level field goal — carries the most information about what these teams actually are.

This is not a prediction. We hold no NFL model. Five 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.

Spreads from market sources as of Sunday morning. Probabilities derived using a 13.5-point standard deviation for the NFL: a 1.5-point favorite wins about 54% of the time, a 3-point favorite about 59%. Where we hold no model of our own, we say so rather than dressing a line up as a forecast.

The Postseason, Where It Stands

Both Division Series are decided. Cleveland won the ALDS after trailing Tampa Bay 0–2 and winning the next three games, including a 2–0 clincher in Game 5 on Saturday. The Rays won the other ALDS series, beating New York 3–0. The ALCS is set: Cleveland against Tampa Bay, two clubs that just played five games, with the Guardians carrying the momentum of a full comeback.

In the National League, Milwaukee beat San Diego 3–1 to advance. The Dodgers beat Atlanta 3–1. Game 1 of the NLCS is today in Milwaukee, where the Brewers earned home-field advantage by finishing with the best record in baseball. The Mets finished 74–88, last in the NL East for the first time since 2003. They were eliminated on September 14. There is no polite statistical frame for 74–88. It is what it is.

Over-Reactions and Under-Reactions

What to Take Home, and the Road Ahead

The week taught two things worth holding onto. First: the right measurement matters more than the interesting one. Issues #192 and #193 both found that the first pass at a question — defense outpaces offense in improvement; new coaches struggle because of the roster — dissolves once you control for the obvious confounders. The boring answer is the honest one.

Second: a probability statement is not a guarantee, in either direction. Texas was 65% likely to win. Oklahoma won. The statement was still correct. Updating your belief after a single game — in either direction — is the main statistical error this paper exists to slow down.

The queue has five pieces ready. The most perishable sit at the front: a Mets postseason-absence scorecard that will be fully live with the NLCS now underway, and the calibration-Wisconsin piece that anchors the methods curriculum. The payroll piece and the inherited-roster analysis are durable. Issue #200 is four issues away. The milestone piece is planned.