Six Issues, Eight Games, and a Model That Underestimated the Mets.
The Week in Review: Issues #37–42
- #37May 4 — Year One Tells You 60% of Everything You’ll Ever Know About a Coach — The stabilization curve for new college football head coaches.
- #38May 5 — The Mets’ Next 17 Games, Mapped — Log5 forecast of every series through May 20. Headlined the issue with a six-win projection.
- #39May 6 — The Jets Have Drafted One Hall of Famer in Fifty Years — And so have twelve other teams. The HoF-harvest distribution across the NFL.
- #40May 7 — Mason Miller Is Spectacular. He Is Not, Yet, the Most Unhittable Pitcher Who Ever Lived — The denominator problem with all-time superlatives.
- #41May 8 — The Jets Are 25th in Triple-Bust Rate — and 10th in Clean-Hit Rate — Two halves of the same draft-evaluation question, decoupled.
- #42May 9 — The Hall of Fame and Hall of Shame of American Sports Owners — Owner-tier rankings across leagues, with the Mets’ ownership change as anchor case.
The Prediction Scorecard · Issue #38’s Log5 Model
Issue #38 used the log5 formula to project the Mets’ next seventeen games against six different opponents, ending May 20. We are now eight games into that schedule. Here is how the model has done, series by series, against the actual outcomes through Saturday May 9.
| Series | Games | Model Said | Actual | Grade |
|---|---|---|---|---|
| @ LA Angels | May 2–3 (2 g) | 0.93 wins (.467) | 1 W, 1 L (lost 4–3, won 5–1) | HIT — within margin |
| @ Colorado | May 4–7 (3 g) | 1.22 wins (.408) | 2 W, 1 L (won 10–5, won 4–2, lost 6–2) | OVER — +0.78 above pace |
| @ Arizona | May 8–10 (2 of 3 played) | 0.55 expected (so far) | 1 W, 1 L (won 3–1, lost 2–1) | OVER — +0.45 above pace |
| Cumulative | 8 games | 2.70 wins expected | 4 W, 4 L | OVER by 1.30 wins |
The model said the Mets would win about 2.7 of these eight games. They won four. That’s a gap of 1.3 wins over an eight-game window — small in absolute terms, but a forty-eight-percent overshoot of the central estimate. The Mets, in this stretch, have been roughly a .500 team. The model said they would be a .337 team.
What This Tells Us About the Model
The log5 model uses run-differential-based true-talent estimates. It said the Mets’ honest level was around .387, below their record-implied .323. Their realized record over these eight games (4–4 = .500) sits even higher. Both numbers are within the wide confidence interval that any 8-game window produces — but the direction of the miss is informative. The Mets’ underlying baseball was less bad than their record looked, and the model knew that. What the model did not know is that even that estimate was still pessimistic, because the run differential it leaned on had been pushed downward by a small number of one-sided losses and a bullpen that, as Issue #40 noted on Thursday, gives up grand slams in the eighth.
Strip those out and you get something closer to a .470 team with awful late-inning luck. That description, on this morning’s evidence, is also closer to the truth than either of the alternatives.
What We Got Right
The Direction Was Correct
Issue #38 said the Mets would lose more than they win in this stretch. They are 4–4 with the Yankees series still ahead. The direction of the prediction is right. The closing sentence — “That is not a recovery pace. That is the pace of running out of road” — remains the right read across the full seventeen games. Going .500 against the Angels, Rockies, and a half-Diamondbacks slate is the floor, not the ceiling.
The methodological point also held. Issue #38 argued that record over a small sample is a noisier estimator than run differential per game, and that the Mets’ true level was somewhere between the two numbers. Eight games in, that is exactly what the data is showing.
What We Got Wrong
The Magnitude Was Pessimistic
The point estimate of 6 wins in 17 was low. At the current pace — 4 wins in 8 games — the Mets project to roughly 8 or 9 wins over the full seventeen, two or three above the model’s headline. The honest framing: the .337 win rate the model produced was the central estimate of a wide distribution, and the actual realized rate is sitting near the upper edge rather than the middle.
The model is not broken. It is being slightly outperformed by a team whose underlying numbers are slightly better than the inputs the model was given. This is the most common kind of model miss: directionally right, magnitude soft. We’ll re-grade in two weeks.
“The Mets going .500 over eight games is not, in any larger sense, good news. It is, statistically, the kind of thing that happens to bad teams roughly thirty percent of the time. The model knew this. The model just didn’t expect this thirty percent.”
— The Sports Page, Sunday Edition No. 005Over-Reactions and Under-Reactions
Over-Reactions
Issue #40’s headline on Mason Miller leaned hard on his slider profile and his 25⅕-inning scoreless streak. Re-reading it Sunday morning, the warning about denominator size landed correctly — but the deck still framed Miller’s present as something close to a generational performance. He’s extraordinary. He is also six appearances and a 34-inning streak deep into a comparison with pitchers who threw 200–plus innings. The piece said all of this. The framing should have leaned harder on the “and we’ll know in 2032” part.
Under-Reactions
The bullpen is still the Mets’ actual story, and Issue #38’s log5 framework cannot see it. The model treats the team as a single quality estimate; it has no way to encode “the starters pitch to a 3.40 ERA, the relievers pitch to a 5.80.” If we had built a separate starter-vs-bullpen projection into the forecast, the Yankees series next week would project worse. That’s methodological work for a future issue.
What Readers Read · May 3–May 9
Readership data pending — the analytics pipeline installed recently, and last week’s counts were not captured. This section will populate starting with the first full week of data.
The Road Ahead
The pipeline is full. Five new pieces drafted Friday are now in the queue: a methods piece on cost-per-win using all thirty MLB teams (the regression’s actual R² is a remarkable .014, which is its own story), a sociological piece on why nobody hates the Colorado Rockies, a 25-year retrospective on the Mets-Braves rivalry, a tactical follow-up on the Mets’ eighth-inning bullpen problem, and a methods follow-up showing three different ways a sports columnist could deceive a reader using the same scatter plot. Plus the existing pieces underneath. There is no shortage of material.
The Mets play the Cubs today, then host Detroit (May 12–14) and the Yankees (May 15–17). The Yankees series is the test of Issue #38’s prediction — it was the only stretch in the projection where the model favored the Mets to lose all three. If they take even one, the model’s pessimism becomes the larger story of this Sunday Edition’s scorecard. If they get swept, the model was right and we just paid attention to the wrong eight games this week.
“Six issues, eight games, one half-resolved prediction. The model was directionally right. The Mets were directionally bad. We will see how the math closes out by May 20.”
— The Sports Page, on the long game and the half-resolved one
