The Mets Are 22–29. The Hurricanes Lost a Game. Memorial Day Weekend.
The Week in Review: Issues #51–56
- #51May 18 — Three Lies, Thirty Dots. A guided tour, in the tradition of Darrell Huff, of how the same payroll-and-wins dataset can be drawn to support three opposite stories without a single false sentence.
- #52May 19 — What the Mets Need from Here Is Not a Miracle. It Is Their Preseason Projection. From 20–26, the road to 87 wins is .578 baseball. PECOTA and FanGraphs had them at that pace in March.
- #53May 20 — Do Patient Teams Win More Games? A reader (Chris Kelly) asks about pitches per plate appearance. The team-level correlation is r ≈ 0.07. Indistinguishable from zero.
- #54May 21 — The Mets Bought Two Closers. Both Are Broken. A $51M closer, a 38-year-old setup man, and a Kimbrel grand slam — with, as it turns out, a substantially wrong dateline.
- #55May 22 — A Five-Game Lead at Season’s End Is Statistical Noise. Here’s the Math. Significance Series, Part 1. χ² > 3.841, and how big the gap actually has to be.
- #56May 23 — Carolina Is 8–0 in the Playoffs. That Has Never Happened in This Format Before. Filed Saturday morning. By Saturday morning, Carolina was already 8–1, having lost Game 1 of the Eastern Conference Final to Montreal on Thursday night.
Six issues. Two methods pieces, one reader question, one Mets pep talk, one Mets pan, one historical NHL marker that aged badly inside forty-eight hours. The pattern of the week is that two of the six are now in the corrections column.
The Prediction Scorecard · Week 7 Resolution
Four predictions from the past week (and one carry-forward Bayesian projection from Issue #45) had something to grade against by this morning. The scorecard is honest, and the scorecard is not flattering.
| From | Prediction | Outcome As of Saturday Night | Grade |
|---|---|---|---|
| #52 May 19 | From 20–26, Mets must play .578 over remaining 116 games (67–49) to reach 87 wins, the average NL third wild-card cutoff. PECOTA’s preseason pace. | Mets went 2–3 in the first five games after the piece ran. Now 22–29, ~7 games back of the third NL wild card (Cubs, 29–23). Week-1 pace is .400, well under the .578 required. | TRENDING MISS — one week of evidence, but in the wrong direction. The hole got deeper. |
| #54 May 21 | The veteran-closer stack (Williams + Kimbrel) is a stacked-failure mode: stacking two former All-Stars increases the probability you have two relievers regressing to their median, not the probability of having one good closer. | Kimbrel DFA’d May 22, the day after the piece ran. Williams now on a 10-appearance scoreless streak, 5 saves in May, 0 ER. The pair as a unit no longer exists; half the prediction’s subject was removed by the team before it could be graded. | PARTIALLY HIT — the structural critique was right (Kimbrel was unfixable). The other half (Williams broken) is, on current evidence, wrong: he is the most reliable Met reliever right now. |
| #56 May 23 | Carolina’s 8–0 playoff start — framed as an active streak entering the Eastern Conference Final. | The framing was stale at publication. Carolina lost ECF Game 1 to Montreal 6–2 on Thursday, May 21, two days before the piece ran. They won Game 2 in OT Saturday (Ehlers, 3:29). Series 1–1. Hurricanes are now 9–1, not 8–0. The historical comp-set math (3 of 4 prior 8–0 starts won the Cup) still stands as analysis. The temporal frame did not. | META-ERROR — not a prediction miss; a publication-timing miss. The historical comp work is unchanged. |
| #45 May 12 (carry) | Devin Williams’s true talent settles in the 2.80–3.50 ERA range (xERA-driven Bayesian projection); the 9.95-then-5.68 April ERA was a small-sample artifact that would migrate toward the mean as innings accumulated. | Williams’s May line: 0 earned runs across the month, 10 K, 5 saves through Saturday. Season ERA still 5.68 because of the front-loaded April; xERA still 3.07. The underlying number has been confirmed across roughly a month of additional work. The reported ERA will continue migrating toward the underlying number as the denominator grows. | HIT — the cleanest call still on the books. The piece said the bad outings were not the pitcher and the good outings would resume. They have. |
| #56 (open) | Implicit forward question from the Carolina piece: does an 8–0 start translate to a Cup, given a tiny historical sample of 3-for-4? | Series is 1–1 entering Game 3 in Montreal Monday night. The Hurricanes are still favored, but the “unbroken” framing is gone. | PENDING — reopens with Game 3. Sunday Edition No. 8 will grade. |
What This Tells Us
One outright HIT (Williams xERA), one PARTIAL (Kimbrel removed before the piece could be properly tested), one TRENDING MISS (Mets wild-card pace), one META-ERROR (Carolina framing stale by 48 hours), one PENDING (the Hurricanes’ actual Cup run). The stat-card headline of 1 of 4 is the honest reading. The more-charitable reading would be 1.5 of 4, counting the structural half of the Kimbrel critique. The Devin Williams call from twelve days ago is doing the heavy lifting; everything else is, this week, somewhere between still-open and walked-back.
What We Got Right This Week
Devin Williams Is the Pitcher Issue #45 Said He Was
Twelve days ago, in Issue #45, the newsletter argued that the “ineffective” Devin Williams of April was a small-sample artifact and that his true talent — visible in his xERA of 3.07 — would migrate the headline ERA back toward where his career-long Bayesian prior already had him. The reported ERA is still 5.68, because the April blowups will sit in the cumulative denominator for a long time. The reported ERA is also no longer the right number to look at. Across all of May, including this past week, Williams has not allowed an earned run. Ten straight scoreless outings. Five saves on the month. The reported ERA is shouting; the underlying number is calm; and the underlying number is the one the Bayesian model has been pointing at since May 5. This is, on the current scorecard, the cleanest still-open call in the project.
Issue #53’s Reader Question Held Up to the 2026 Standings
Chris Kelly’s question about whether patient teams win more games produced a piece that said, plainly, the team-level correlation between pitches-per-plate-appearance and winning percentage is r ≈ 0.07 — not different from zero in any practical sense. As a check, today’s standings: the Astros, year after year one of the most patient teams in baseball, sit at 18–31 in the AL West. The Pittsburgh Pirates, a free-swinging organization, are over .500. Cubs (patient) and Tigers (free swingers) are both running away with their divisions. The standings are not, in fact, sorting by P/PA. The piece was, as Chris’s eye half-suspected, correct in the direction it pushed back.
Issue #51’s Methods Held
Three Lies, Thirty Dots was a teaching piece, not a prediction. It has no scorecard entry. It does, however, deserve a green-box mention this week: of the four ways to read the payroll/wins scatterplot it walked through, the honest reading (R² = .014, no signal) is the one the standings have continued to support across the month since the underlying dataset was assembled. The Marlins still lead the league in cost-per-win and are still 19–30. The Dodgers are still expensive and still very good. Nothing in between. The piece’s claim that payroll explains essentially none of this year’s wins remains true.
What We Got Wrong This Week
The Mets Wild-Card Math, Week One
Issue #52, published Tuesday, was not wrong on its math: from 20–26, the road to an 87-win wild-card spot is .578 baseball over 116 games, and PECOTA had the Mets at exactly that pace in March. The piece’s implicit assumption — that the projection-implied pace was achievable in week one — is now under pressure. The Mets played .400 ball over the five graded games and lost ground in the wild-card chase. The third NL wild card belongs to the Cubs at 29–23; the Mets are roughly seven games behind that bar and falling. One week is not a verdict. It is a data point in the wrong direction, and it sits on top of an injury situation (Holmes out, Lindor not yet running, Alvarez until mid-July, Senga uncertain) that the model in Issue #52 could not encode. The verdict comes by the All-Star Break. Until then, the call remains live, and the Sunday Edition will keep grading it.
The Hurricanes’ Streak Was Already Broken
Issue #56, published Saturday morning, treated Carolina’s 8–0 playoff record as an active streak entering the Eastern Conference Final. The streak was, by then, 8–1: Carolina lost ECF Game 1 to Montreal 6–2 on Thursday, May 21, with Juraj Slafkovsky scoring twice and Jakub Dobes turning in a 25-save win. The Canadiens put up four goals in the first period. By the time the issue published Saturday morning, Carolina had also already won Game 2 in overtime (Ehlers at 3:29). The series, at issue release, was 1–1. The piece’s historical work — the four prior 8–0 starts, the 3-of-4 Cup conversion rate, the “Brind’Amour’s analytics check finally clearing” framing — is unchanged by the loss. The comparison to 1985 Edmonton still stands. The temporal frame did not. The mistake was operational: a piece written from data that ended Tuesday, run on Saturday, with no morning-of check against the news. Fix: queue items older than four days will require a freshness review before they auto-publish.
The Issue #54 Dateline
Issue #54 (“The Mets Bought Two Closers. Both Are Broken,” bylined The Heckler, published Thursday May 21) opened with the phrase “last night,” in reference to Craig Kimbrel surrendering a grand slam to the Rockies’ Jake McCarthy in the eighth inning of a 2–2 tie. The Mets lost the game 6–2. All of these facts are true about a real game. None of them are true about the night of May 20, 2026 — which is what “last night” meant in the piece. The actual May 20 game was a Mets 8–4 loss at Washington. The Coors grand slam happened on the May 4–7 road trip, roughly two weeks before the column was published. The error is, in absolute terms, not catastrophic: the structural critique of the bullpen still landed (and the team agreed, designating Kimbrel for assignment the next morning). The error is, in editorial terms, a real one: a column dressed a real event from two weeks earlier in the costume of breaking news. Every game-anchored piece from here forward will require an explicit date check before it leaves the queue.
“The Mets are six games out of last place and going the wrong way. Carolina’s streak ended in a four-goal first period. The job is to write what the math says about both, plainly, on the Sunday after.”
— The Sports Page, Sunday Edition No. 007Over-Reactions and Under-Reactions
Over-Reaction
The bullpen piece (Issue #54) over-reached on the “both broken” framing in two specific ways. First, the dateline error already covered above. Second, and more substantively: by Thursday morning Devin Williams had already strung together a multi-week scoreless run that the piece does not engage with. The Heckler had a real point about the structural risk of stacking veteran closers, and the team validated half of it (Kimbrel DFA’d Friday). The piece would have been stronger for acknowledging that one of the two closers had, by publication time, stopped being broken. The math was generous to a sharper take than the column delivered.
Under-Reaction
The Hurricanes piece (Issue #56) under-reacted to the obvious question, which is now the only question that matters: what happens if 8–0 becomes 8–3 in a hurry? Game 1 in Raleigh against Montreal was not a close loss. It was 6–2 with four Canadiens goals in the first period. The piece’s historical comp set (1952 Detroit, 1960 Montreal, 1969 St. Louis, 1985 Edmonton) is exactly the right framing for what an 8–0 start has historically meant. It has nothing to say about what a team does when the run finally ends in a six-goal loss to a Slafkovsky-Suzuki-Caufield top line. That is the question Sunday Edition No. 8 will have an answer to. This one does not.
What Readers Read · May 17–May 23
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 · Week 8
Two pieces are planned to anchor the coming week, with the rest of the queue rotating around them as the variety rule allows.
Significance Series, Part 2 — The Tiebreaker Trap. Issue #55 introduced the chi-square test and showed that a five-game lead at season’s end is statistical noise; an 80-point gap in winning percentage across 162 games is the minimum the test will recognize. Part 2 takes the next obvious case: what happens when two teams finish dead even and the league’s tiebreaker apparatus — head-to-head, common opponents, division record — promises to settle a question that statistics cannot. The 2024 Mets–Braves 89–89 tie is the working example. The piece argues, with the chi-square in hand, that the tiebreaker rules are a coin flip in formal dress.
The Mets Pythagorean Audit. A companion to Issue #52’s wild-card piece, asking the question Bill James spent forty years asking: are the Mets’ 22–29 a fair reflection of how they have been playing, or has the run differential told a different story? Pythagorean expectation, applied to the current data, will either reinforce or revise the .578-required-pace argument. Either way, it is the next honest test of the model in Issue #52.
Beyond those: a Notre Dame recruiting piece is ready in the queue and will run when the variety rule allows a CFB beat; a piece on the Mariners’ Dipoto-era 54% strategy waits its turn; the BaseRuns Mets diagnostic and the Mets Misery Index NY-vs-DC piece both stand by. The Hurricanes’ Game 3 at Montreal is Monday night, and a follow-up will be queued if the result demands one.
“Fifty-seven issues in. Four hundred forty-three to go. We will keep checking the work, on Sundays, with the misses written in plain ink. That is still the deal.”
— The Sports Page, on the long game
