We Ranked Them 33rd. They Beat Kent State 57–0.
The question we ask every Sunday: what did we claim this week, and did reality agree?
The scoreboard. Six issues, four gradeable predictions: two hit, zero missed, one partially confirmed, one still pending. Three pieces made no forward forecast at all — which is worth saying out loud rather than quietly skipping.
Here is the surprising part. There is no scoreable miss this week. We looked. Two calls came in cleanly: Tampa Bay holds a 3.5-game AL East lead (we said 72.1% probability of winning the division), and the Mets’ path to the playoffs collapsed precisely the way the math said it would. The South Carolina dark horse call is still open — a 57–0 win over Kent State is a result, not a verdict, and they remain unranked — and the season is 12 games long. We will not manufacture a miss. If the misses were dull, say so.
Why we bother: a prediction nobody grades is just a story. You deserve to know which stories held up.
The Week in Review: Issues #156–161
- #156Aug 31 — The Mets Cannot Ruin Anybody. The Athletics Can Ruin Four Seasons. — Which eliminated teams still control playoff fates — and Oakland controls more than anyone.
- #157Sep 1 — We Built 252 Ways to Predict the Playoffs. The Best One Lost to the Simplest One. — A 252-metric grid search, leave-one-season-out validation, and a lesson about when to stop tuning.
- #158Sep 2 — The Mets Must Pass Seven Teams. They Play Two of Them. — Championship leverage on every remaining game. Even a 25–8 heroic run gave ~2.4% playoff probability.
- #159Sep 3 — How Many Stars Does It Take to Flip a Team? Baseball: Four. Basketball: One. — WAR additivity, roster depth, and why sport structure explains the difference.
- #160Sep 4 — Two Teams Have Everything the Contenders Have. They Are Ranked 33rd and 35th. — CFB dark horse analysis: South Carolina and Florida, ranked outside the top 25, with returning production metrics that say eight wins.
- #161Sep 5 — The Best Season to Watch Is Not the Best Season to Have. — A hope index for all 30 MLB clubs. The Mets scored 1.1 out of 100. The null result is the finding.
The Prediction Scorecard
| From | Prediction | Outcome | Grade |
|---|---|---|---|
| #153 Aug 21 |
Tampa Bay wins AL East — 72.1% probability at time of writing | As of Sep 6: 83–57, leading AL East by 3.5 games, magic number 20. On pace to clinch before month’s end. | HIT — direction and probability both validated |
| #158 Sep 2 |
Mets’ playoff ceiling of ~2.4% even with a heroic 25–8 run, given remaining schedule structure | Mets are 64–77 as of Sep 6. They would need to go 34–5 in their final 39 games to reach .500. Effectively eliminated. | HIT — the math was more optimistic than reality turned out to be |
| #160 Sep 4 |
South Carolina deserves eight-win caliber treatment — returning production metrics rank them in the 30s behind teams already in the top 25 | Beat Kent State 57–0 in Week 1. Beat Kent State 57–0. Still unranked in the first in-season AP poll; Massey’s composite has them 34th and falling. | PENDING — one blowout over a mid-major decides nothing |
| #160 Sep 4 |
Florida deserves eight-win caliber treatment — ranked 35th by returning production, unlisted in the AP preseason poll | Beat Florida Atlantic 66–21 in Week 1. Still unranked in AP. Too early to judge. | PENDING — check back in four weeks |
Issues #157, #159, and #161 made no forward predictions — methods retrospective, structural analysis, and a null-result study respectively. Not every issue needs to bet on the future. But we count them so the ratio stays honest.
What We Got Right
The Wins
The Mets schedule analysis was our clearest hit. Notice what the model actually said: not that the Mets would fail, but that the structure of their remaining schedule made the climb nearly impossible regardless of how they played. They had to pass seven teams and faced only two of them directly. The math gave them a 2.4% ceiling on a heroic run. They are currently 13 games below .500. The ceiling was generous.
The Tampa Bay call has been tracking since August 21. We said 72.1% probability of winning the AL East, based on their then-record and remaining schedule. They have won eleven of their last fifteen and now hold a 3.5-game lead with 22 to play. This one still has three weeks to go wrong — a magic number of 20 is not a championship — but the direction is right and the margin is comfortable.
South Carolina surprised us in how fast it resolved. We published the dark horse piece on Thursday. By Saturday they had shut out Kent State 57–0. The voters were unmoved: still unranked when the first in-season poll appeared on Tuesday. We want to be careful here: one blowout over a mid-major program is not a validated season. But the preseason undervaluation was real, and the first data point confirmed it.
What We Got Wrong (or Are Still Watching)
The Honest Accounting
There is no scoreable miss this week. That is not a boast — it is a description, and you should hold it lightly. Three of six issues made no forward predictions at all. A 2-for-2 week (plus one PENDING) with three abstentions is not a perfect week; it is a quiet one. We were not bold enough to miss.
The overfit grid piece (#157) is the honest case study in intellectual humility. We built 252 metrics, optimized them in-sample, and found the best one — 59.2% accuracy — fell apart out-of-sample to 48.9%. Raw run differential at 57.8% beat the tuned model. We published that result. Publishing a null finding is harder than publishing a success, but the Sunday Edition exists for this reason: the null findings are the teaching moments.
Florida is the open question. We said they deserved eight-win caliber treatment. They beat a mediocre FAU program by 45 points. That proves nothing about their SEC schedule. We will revisit in four weeks, when they have played two or three conference games.
“The overfit lesson is not that models fail. It is that they fail in a specific way: they memorize the past instead of learning from it. You now know the difference. That’s worth something.”
— The Sports Page, Issue #157Over-Reactions and Under-Reactions
Over-Reactions
We may have over-weighted the South Carolina Week 1 result in real time. A 57–0 win over Kent State tells you South Carolina’s offense is functional and their defense can dominate a lesser program. It does not tell you how they fare in a conference game against Georgia or Tennessee. The returning production metrics that placed them in the 30s behind top-25 teams were built on multi-year data. One blowout is three hours of evidence. Hold the champagne.
Under-Reactions
The hope curve null result (#161) deserves more attention than it got. We found p=0.055, r=0.086 — not significant, barely trending. Our conclusion was correct (no relationship between hope curve shape and playoff success). But the finding cuts deeper than that: fans might be right that a rollercoaster season feels different, without that feeling predicting anything about outcomes. The statistics are not the whole story. We said this in passing and should have dwelled.
What Readers Read · Aug 30–Sep 5
Readership data pending — the analytics token was not available in this publishing environment. This section will populate in the next edition.
What to take home, and the road ahead
You can do something now that you could not do a week ago. You can look at a CFB preseason ranking and ask: what is this based on? If the answer is reputation and last year’s record, you know it may be wrong in the same direction South Carolina’s was — undervaluing teams that replaced key pieces through smart recruiting and development. You can ask about returning production. That is the metric the AP poll cannot easily quantify.
You can also look at the Mets’ season differently. Not as a collapse, but as a calibration check. The schedule leverage math said the path was essentially closed in early September. The Mets are now 13 games under .500. The math was not pessimistic — it was accurate. Sometimes the numbers just tell you what the standings already show.
Next week’s pipeline opens with the payroll arms race in MLB (why throwing money at pitching is not working the way teams think), then the “one that got away” retrospective on the Mets’ 2015 window, and the Week 1 signal study — how much does a single week of college football results actually predict? That last one is exactly what you need right now, while the South Carolina result is still fresh.
“Twenty-two Sundays in. The scorecard is honest: more hits than misses, more abstentions than we probably should have. We’ll fix that by being bolder. You’ll hold us to it.”
— The Sports Page, Sunday Edition No. 022Correction · 8 September 2026
The first version of this issue, published on 6 September, said South Carolina had “gone from unranked to No. 10 in the AP poll.” That was wrong, and it was the headline.
No such poll existed when we wrote it. College football’s first in-season AP poll had not been released — week one ran through Labor Day, so it did not appear until Tuesday 8 September. When it did, South Carolina was not in it. Texas A&M is No. 10. Massey’s composite of fifteen rating systems has South Carolina 34th, down three places from preseason.
The 57–0 win over Kent State is real and was verified. The ranking attached to it was not. We graded our own dark-horse call a partial hit on the strength of a poll that had not happened; it is now marked PENDING, which is what it always was.
How it happened, since that is the useful part. The issue was assembled at 4:17am and asked a web search what the new poll said. Search summaries will answer that question whether or not the poll exists, reconstructing a plausible ranking from a previous season. Every fact around it was true — the score, the opponent, our original 33rd-place ranking — which is exactly what made the one false link convincing.
We have since added a rule: no forecast is graded against a poll, award or ranking before its scheduled release, and nothing is graded on a source we cannot confirm from a primary feed. A newspaper about bad inference does not get to make this particular mistake quietly.
