The Sports PageWe Made No Bets This Week. The Tigers Proved Our Point Anyway.
The question we ask every Sunday: what did we claim this week, and did reality agree?
The scoreboard. Six issues, no forecasts. Nothing hit, nothing missed, nothing to account for. All six were methods, history or measurement — not bets.
Here is the surprising part. A quiet scorecard is not a quiet week. Oklahoma State beat Oregon 39–31 as a 23-point underdog — about a 1-in-13 result. Thirteen of those sit on one Saturday card, which is why one lands almost every week.
Why we bother: a prediction nobody grades is just a story. When there is nothing to grade, we say so in one line and teach you something instead.
The Week in Review: Issues #163–168
- #163Sep 7 — Rutgers Was Favored by Twenty-Nine. It Lost by Sixteen. — The hardest opening week to predict in three years: 47 rated games, 14.3 points per game off the SP+ number. 28% worse than the prior two seasons combined.
- #164Sep 8 — 209 of Every 1,000 Never Pitch Again. — Why the claim that teenagers are 57% of all Tommy John patients tells you nothing about your son’s arm. A denominator lesson for any percentage you are handed.
- #165Sep 9 — One Game Told Us More Than the Whole Summer Did. — Recruiting rankings and returning production explain 14% of how a college season turns out. Add one Week 1 result and it climbs to 21%. We expected the opposite.
- #166Sep 10 — Every Team Has Its Babe Ruth. The Mets Have Two. — Nolan Ryan for Jim Fregosi in 1971. Pete Crow-Armstrong for a two-month shortstop in 2021. The same trade, fifty years apart. The cost framework that explains both.
- #167Sep 11 — The SEC Has an Edge. It Is 0.4 Points. — The conference premium is real, measurable, and smaller than anyone argues about it. Notre Dame ranked 22nd in schedule difficulty, harder than 46 of the 67 power-conference slates. We will take that one on the chin.
- #168Sep 12 — Detroit Outscored Its Opponents by 55 Runs. It Is Twelve Games Under. — Year-to-year correlation of one-run records: near zero. A team that plays poorly in close games this year is not a team that plays poorly in close games next year. Detroit’s collapse is noise wearing a record.
The improbable thing that happened yesterday
Oklahoma State beat Oregon 39–31. Oregon was favored by 23¼ points.
Turn that spread into a probability and Oklahoma State had about a 7.9% chance — call it 1 in 13. Our own ratings, frozen before kickoff, were slightly kinder to them: Oregon by 20.3, or 1 in 9.
So: a stunner; now here is the part people get wrong.
There were 49 rated games on that card. Add up every underdog’s chance of winning its own game and you get 11.4 — the number of upsets a normal Saturday should produce. The actual count was 15.
Notice what that does to the word "improbable." A 1-in-13 shot is unlikely in its own game; across a full card it is close to inevitable that something like it happens. The surprise was never that an upset occurred. The surprise is only ever which one.
This is the mistake behind every "nobody saw this coming" segment you will hear today. Somebody did; thirteen somebodies, roughly, each wrong about which game.
| Yesterday’s least likely winners | Score | Chance |
|---|---|---|
| Oklahoma State beat Oregon | 39–31 | 1 in 9 |
| Georgia State beat Kennesaw State | 31–17 | 1 in 7 |
| Florida Atlantic beat Navy | 38–30 | 1 in 5 |
| North Texas beat UNLV | 44–6 | 1 in 5 |
Chances from our ratings frozen on Thursday, before any of these were played. Grading an upset on ratings updated after it is how you convince yourself you saw it coming.
What to watch today, and why
The best game on a card is rarely the one with the best teams; it is the one nearest a coin flip, because that is where the result carries the most information.
| Closest games | Line | Favorite wins | Kickoff |
|---|---|---|---|
| BUF at HOU | BUF by 1.5 | 54% | 1:00 PM EDT |
| NYJ at TEN | TEN by 1.5 | 54% | 1:00 PM EDT |
| GB at MIN | MIN by 2.5 | 57% | 4:25 PM EDT |
| CHI at CAR | CHI by 3.0 | 59% | 1:00 PM EDT |
| DAL at NYG | DAL by 3.0 | 59% | 8:20 PM EDT |
Probabilities from the closing spread with a 13.5-point standard deviation. We hold no NFL forecast of our own yet — two weeks of data is not a model, and we would rather say that than dress a guess up as one.
The Prediction Scorecard
Issues #163–168 made no forward predictions — methodology, historical analysis, and retrospective review. One carry-over from Issue #143 remains open.
| # | Prediction | Outcome | Grade |
|---|---|---|---|
| 163 | CFB Week 1 was measurably the hardest to predict in three years; this measurement will be run after every week. | Week 2 data not yet compiled. Tracking commitment, not a scoreable forecast. | N/A — methodological commitment |
| 164 | No forward prediction made. Piece was a denominator lesson. | — | N/A |
| 165 | Week 1 results explain ~21% of final-season outcome versus 14% without; informative but one club in four will mislead you. | Methodological finding; grades against December standings, not Week 2. | PENDING — season-long check |
| 166 | No forward prediction made. Piece was historical retrospective. | — | N/A |
| 167 | No forward prediction made. Piece was schedule-strength measurement. | — | N/A |
| 168 | Detroit’s one-run record is noise; the pattern will not repeat next year. Correlation near zero across all clubs, all seasons. | Within the week: Tigers went 4–2, losing two of three one-run games. Differential climbed from +55 to +63. The pattern continues in-season, consistent with the model. | PENDING — year-over-year check (2027) |
| 143 carry-over |
Florida (ranked 33rd) and South Carolina (ranked 35th) are better than their preseason rankings suggest. Model projects 7.9 and 7.8 wins respectively out of twelve. | South Carolina is 2–0 after beating Kent State 57–0 and Towson 45–9. Two wins over non-power opponents. The ranking jury is not even seated yet. | PENDING — ask again in October when SEC play begins |
What We Got Right
The Real-Time Validation Nobody Ordered
On Friday we published a piece saying that Detroit’s one-run record is luck, not a close-game flaw. The club is 69–79 having outscored its opponents by 63 runs, which projects to about 81–67. Its one-run record is 14–27. Year-to-year correlation of one-run records across the league is near zero.
Then they went 4–2 on the week. Three games were decided by one run and they lost two of them; the others they won by 3, 5, 4 and 4. Six games proves nothing. It is simply the shape the piece described: comfortable wins, coin-flip losses.
You cannot plan for validation that comes the next day. When it happens, you note it and keep going. The real check is whether the 2027 Tigers carry this year’s one-run record forward into next season. They won’t. That is the prediction.
Week 1 Signal, Two Weeks Later
Issue #165 found that a single Week 1 result adds seven percentage points of explanatory power to preseason ratings (from 14% to 21%). Notre Dame went from No. 4 to No. 3 after a dominant Week 1 win over Wisconsin at Lambeau Field, then beat Rice 52–0 in Week 2. Oregon, which drew suspicion after a close Week 1, fell to Oklahoma State in Week 2.
This is not a vindication of the model — it is two data points. But neither contradicts the finding, and the finding was that Week 1 is more informative than anyone expected going in. Hold it loosely. December is when you grade it.
What We Got Wrong (or Did Not Say)
The Mets, in Plain Terms
The Mets are 64–78 as of this morning. They are not close to a wild card spot. The two-trade retrospective in Issue #166 was about the historical cost of trading Ryan and Crow-Armstrong — not a forecast for 2026. But we have been covering this team all year, and the honest thing to say is that we never made a formal record prediction early in the season, so there is nothing to grade.
What we can say: they are exactly as disappointing as Pythagorean expectation would have let you expect given their performance. The surprises were not hidden in the math. They were hidden in the lineup.
The Babe Ruth piece does not become wrong because the 2026 Mets are 64–78. The trades it describes were wrong the day they were made, on any accounting that looked past the current roster.
Nothing This Week — and Why That Is Worth Noting
A week with no forward predictions is a week with no accountability surface. You cannot miss a call you did not make. There is real value in methodological pieces — Issue #164’s denominator lesson and Issue #167’s schedule-strength study will be useful to you for years — but the ratio of predictions to analysis is something to watch. We said we would be bolder. This week we were not.
Coming Monday: a piece that admits our backtesting model cannot beat the market. 51.0% across 2,140 games, confidence interval contains 50. That is not a method; that is noise in a spreadsheet. We will say so in the headline.
“A week with no predictions is a week with no accountability surface. You cannot miss a call you did not make. Notice when that feels like safety.”
— The Sports Page, Sunday Edition No. 023Over-Reactions and Under-Reactions
Over-Reactions
Notre Dame beat Rice 52–0 on Saturday night. Notice how fast the takes arrived. Rice. Notre Dame is No. 3 in the country and Rice is one of the worst teams in a rebuilt conference. The Week 1 signal piece said openers are informative because they pit competitive opponents against each other. A Week 2 home blowout against a 110th-ranked opponent is not an informative game. Hold the coronation.
The same caution applies to South Carolina at 2–0. Both wins were over teams from the lower half of the FCS and Group of Five landscape. The dark-horse prediction from Issue #143 is about December, not about shutting out Kent State in August.
Under-Reactions
Oregon fell to Oklahoma State in Week 2 — and most of the discourse treated it as a fluke. Look at the Week 1 issue again. The whole point of that measurement was that the hardest-to-predict opening week in three years suggests something about the calibration of preseason ratings, not just individual game variance. When a highly ranked team loses in Week 2, it is worth asking whether Week 1 was already telling you something.
The MLB playoff picture deserves more attention than the pennant-race coverage it is getting. The Brewers are 92–56 and appear to be running away with the NL Central. That is one of the better records in baseball. You have probably read three articles about the Braves this week and none about Milwaukee.
What Readers Read This Week
Readership data unavailable this edition — the analytics API was unreachable at publish time. The archive is at thesportspage.net.
Correction · September 13, 2026
The first version of this edition, published this morning, said Detroit "went 3–4 with all four losses by one run," and later that they "lost three games by one run and won two" in a four-game series. Both were wrong, and they contradicted each other.
From the MLB Stats API: Detroit went 4–2 between September 6 and 12. Three games were decided by one run and they lost two of them. We also put their Pythagorean projection at 84–62; on a +63 run differential it is about 81–67, against an actual record of 69–79.
The point of Friday’s piece survives, which is exactly why the error is worth naming: a number that flatters your argument is the one you are least likely to check.
What to Take Home, and the Road Ahead
The week ahead opens with something we are genuinely curious about. The returning-production piece asks whether Penn State returning 7.6% of its offense while winning 45–0, and Clemson returning 41% while losing by 41, is a signal or a coincidence. We ran eleven seasons of data. The answer is “mostly coincidence,” and it is more interesting than the alternative would have been.
Later in the week: log-odds. You have seen the home-field advantage figure, the single constant of +0.14. That constant is the whole reason the log is there — percentages cannot hold a constant and have it mean the same thing across the full range of team quality. This is one of those methods pieces that sounds dry and changes how you read a probability for the rest of your life.
And then — the one to read twice — No Edge. A perfect backtest is a bug report. Rebuilt so the method could not see the answers it was trying to predict, it beat the spread 51.0% of the time across 2,140 games. The 95% confidence interval contains 50. We will say in the headline that we were wrong, and we mean it.
CFB Week 3 starts Saturday. Notre Dame hosts a real opponent for the first time. South Carolina opens conference play. The dark-horse prediction starts generating real evidence. Watch for it.
“Twenty-three Sundays in. The ratio of analysis to prediction has been running high. We know. We will fix it by being bolder — and then we will hold ourselves to that in public.”
— The Sports Page, on the long game
