Fifty Issues, Fifty Lessons, and One Federal Coinage We Repurposed for Ourselves.
The Week in Review: Issues #44–49
- #44May 11 — The Jets Are the NFL’s Second-Worst Drafters in Rounds 1–3. The Worst Are the Patriots. The Patriots Won Six Super Bowls.
- #45May 12 — The Mets Paid $51 Million for the 0.33-ERA Devin Williams. He Was Never Going to Be 0.33-ERA Devin Williams.
- #46May 13 — Notre Dame Took Seven from the Portal. One Thousand Two Hundred Other Players Have Nowhere to Go.
- #47May 14 — The Rockies Have Never Won the West. Nobody’s Mad About It.
- #48May 15 — The Cost of a Win, and Why That Sentence Is Mostly a Lie.
- #49May 16 — Seventeen to Two: The Rivalry Where One Side Stopped Showing Up Years Ago.
Six issues. Two Jets, one Mets bullpen, one Notre Dame portal, one Rockies sociology, one ratio-statistics methods, one twenty-five-year rivalry retrospective. A normal week for a newsletter that does not, by editorial commitment, have a beat.
The Prediction Scorecard · Final Resolution of Issue #38’s Log5 Model
The big open prediction from the past two weeks closes today. Issue #38, published May 5, used the log5 formula to project the Mets’ next seventeen games against six different opponents. By this morning all seventeen games should be in the books. Here is the full resolution.
| Series | Games | Model Said | Actual | Grade |
|---|---|---|---|---|
| @ LA Angels | May 2–3 (2 g) | 0.93 wins | 1 W, 1 L | HIT |
| @ Colorado | May 4–7 (3 g) | 1.22 wins | 2 W, 1 L | OVER — +0.78 |
| @ Arizona | May 8–10 (3 g) | 0.82 wins (.273/g) | 1 W, 2 L (won 3–1, lost 2–1, lost 6–2) | HIT — +0.18 above projection |
| Issue #45: Williams ERA | May 12–ongoing | ERA settling 2.80–3.50 (xERA-driven) | ERA 5.68, xERA 3.07; 6 straight scoreless outings | PARTIALLY HIT — xERA in range; actual ERA still converging from early blowups |
| vs Detroit | May 12–14 (3 g) | 1.08 wins | 3 W, 0 L (10–2, 3–2 (10th), 9–4) | OVER — +1.92, the model’s biggest miss |
| vs NY Yankees | May 15–17 (2 of 3 played) | 0.76 wins (~0.50 through 2 g) | 1 W, 1 L (lost 5–2 Fri, won 6–3 Sat; game 3 this afternoon) | HIT through 2 g — game 3 today closes the window |
| @ Washington | May 18–20 (3 g) | 1.60 wins | not yet played (Mon–Wed) | PENDING — outside the Sunday recap window |
| Cumulative thru Sat May 16 | 13 of 17 games | 4.6 wins expected | 8 W, 5 L (.615 pace) | MISS — model projected 6.41 total; Mets already at 8 with 4 games to play |
What This Tells Us — verdict updated May 17
The model has been directionally correct, magnitude wrong. Through 13 of 17 games, the Mets are at 8–5 — 3.4 wins above the 4.6 the model expected through this point. The model’s 6.41-win headline for the full 17-game window has already been exceeded. At their current pace, the Mets finish this stretch at 9 or 10 wins. The model said 6. The grade is MISS. The Tigers series, where the model expected 1.08 wins and the Mets swept 3–0, produced nearly two full wins of unexplained surplus on its own. The Yankees series, through two games, has run roughly as projected — 1–1 against an expected 0.50 wins per two games, a near-hit. The model correctly identified the Yankees as the most difficult series; it simply lost all its margin in Detroit.
The deeper story is not the model’s magnitude. It is what happened between Issue #38 and this Sunday Edition: the Mets lost their best 2026 starter to a fractured fibula. The log5 framework treats team quality as a single stable estimate; it has no way to register that Clay Holmes is now on the injured list for 6–8 weeks. Whatever talent estimate the model was working with is the wrong estimate for the rest of the May 18–20 Washington series and everything that follows.
Breaking · The Holmes Fibula and the Mets’ Rotation Crisis
On Friday night, in the fourth inning of the Subway Series opener, Yankees outfielder Spencer Jones drove a 111.1-mph comebacker off the right leg of Mets starter Clay Holmes. Holmes, in a piece of stoicism that has aged badly, stayed in the game and faced seven more batters before exiting in the fifth. X-rays after the game showed a fractured fibula. The expected absence, per manager Carlos Mendoza, is “a long time” — weeks at minimum, months as a real possibility.
Holmes, before the injury, was 4–3 with a 1.86 ERA in his first season as a starter after years as a Yankees reliever. He was, in the cold language of beat reporting, “one of the few bright spots for the last-place club.”
The Mets’ injured list, as of this morning: Holmes (fractured fibula, 6–8 weeks), Kodai Senga (lumbar spine inflammation, targeting a live bullpen today), Francisco Lindor (calf strain, no timetable), Jorge Polanco (Achilles/wrist), Luis Robert Jr. (back), and Francisco Alvarez (torn right knee meniscus, surgery May 14, eight weeks). Four regulars and two starters. The team Issue #38’s log5 model was projecting in early May is not the team that takes the field for the rest of the month.
A Methods Retrospective
Fifty issues is, statistically, a respectable sample. Not enough to make claims about the whole project — the newsletter is aiming for 500, and 50 is ten percent of that — but enough to see the patterns the work has been drawn toward. Six themes have recurred across the first half-hundred issues. Each of them is, in its own way, a way of saying the same thing.
1. Small samples lie. Loudly. With confidence.
The newsletter opened with Skenes’s 67.5 ERA after ⅔ of an inning, which is the cleanest possible example of how a denominator of less than one can produce numerators that read like all-time disasters. Half a dozen subsequent issues have made the same point with different denominators: 5 plate appearances batting .667, 6 games at .500, 31 games at 10–21, 8 games through Issue #38’s log5 window. The recurring lesson, in different costumes: most of what looks meaningful in early-season baseball is the variance of a small denominator playing tricks on the eye.
2. Ratios are dangerous when both terms vary.
This week’s methods piece on cost-per-win (Issue #48) put a name to a structural problem the newsletter had been circling for forty-seven prior issues. Payroll-per-win, runs-per-game in a small sample, ERA over fewer than fifty innings: any ratio where both numerator and denominator move can be optimized by minimizing either, even when the optimization tells you nothing about quality. The Marlins lead Major League Baseball in cost-per-win and are 19–23. The ratio rewards their underspending, not their winning.
3. Stabilization takes longer than fans want it to.
Strikeout rates stabilize around 70 plate appearances. Batting averages take 500 PA. Coaching ability takes a full college football season to be 60% legible. Mason Miller’s “most unhittable ever” case (Issue #40) was bounded by the same problem: 34 innings is not nothing, but it is not the 200-plus inning sample that the comparison required. The newsletter has, with some consistency, told readers to wait. The waiting is usually the right advice.
4. Bayesian recovery is a useful frame.
Several of the newsletter’s cleanest hits have come from Bayesian models that combined a player’s career prior with the small-sample event in front of us. Skenes’s recovery was modeled this way (Gamma-Poisson). Soto’s calf injury time-to-return was modeled this way (Beta-Binomial on prior calf-injury data). The Mets’ 2026 win projection used a similar frame at the team level. The technique is not magic. It is just a more honest accounting for what we knew before the strange thing happened.
5. Folk theories about “why teams win” usually describe an individual-level effect.
The thread underneath the rivalries piece, the owner index, the hate ranking, and the upcoming reader-question piece on pitches per plate appearance: when a sports column tells you that clutch hitting or plate discipline or veteran leadership explains a team’s standings, the explanation is usually doing real work at the player level and almost no work at the team level. The aggregation is where the signal dies.
6. We have been wrong, and we have said so.
Sunday Edition No. 1 owned an overcalled framing on the $467M Mets infield. Sunday Edition No. 4 admitted Skenes’s recovery had been slower than the model called for. Sunday Edition No. 5 noted that Issue #38’s log5 magnitude was, by Day 8, looking pessimistic. The misses are not embarrassments. They are the data we trade in. A model that hits all of its calls is either trivially calibrated or lying about what it predicted.
What We Got Right This Week
Issue #1’s Model Call, Finally Closed
Paul Skenes has a 1.31 ERA in 2026 through nine starts and 58 innings. Issue #1 of this newsletter, published Opening Day, used a Gamma-Poisson model to argue that Skenes’s 67.5 ERA after ⅔ of an inning would dissolve quickly and that his true talent was closer to his career 1.97 ERA. After eight starts at 1.31, the model has been confirmed. Small denominators lie. We said so on March 27. That is now the most time-tested call in the newsletter’s 50-issue run.
The Bullpen Was the Story All Along
Issue #45, on Devin Williams, argued that the $51 million the Mets paid for the “0.33-ERA Devin Williams” was a bet on a regression that was never going to hold. Williams is now at a 5.68 ERA but an xERA of 3.07 — his underlying metrics have settled squarely in the 2.80–3.50 range the Bayesian model projected. The piece said the 9.95 ERA would not hold; it did not. The piece said the true-talent range was 2.80–3.50; xERA agrees. The actual ERA is still above that range because of the small-sample early disaster, but the direction is confirmed. This is a PARTIALLY HIT we will close out in a future Sunday Edition when the sample is large enough to grade fully.
Issue #48 (Cost of a Win) said that payroll explains roughly 1.4% of the variance in 2026 wins through this point in the season. Nothing about this week changes that: the Mets, second-highest payroll in baseball, are in last place in their division with a starting rotation built around a $250 million-plus roster that just lost its best starter to a comebacker.
What We Got Wrong This Week
The Tigers Series Was an Outlier the Model Could Not See
Issue #38’s log5 model projected one win in three games against the Tigers. The Mets won three. That is, by a clear margin, the biggest single-series miss in the model’s scoring window. The model was working from a true-talent estimate based on run differential through May 4, which placed the Mets at .387. A team that goes 3–0 against a .528 opponent over three games is not playing like a .387 team in that window; it is playing like a roughly .700 team, which is what the Mets did against Detroit.
This is not the model being wrong. It is the model being limited in the way log5 frameworks are always limited: a single talent estimate cannot capture three-game stretches where the right starters happened to line up against the right opposing lineup on the right days. The honest read is that the Mets, in their non-Yankees baseball this past week, played significantly above their season talent. The Yankees series, through two games, was a rough wash — 1–1 against an expected 0.50 wins, which means the model nearly called it correctly. The entire surplus from this 13-game window came in Detroit.
“Fifty issues. The federal government called this year a five-morpheme thing for a country’s 250th birthday. We are doing the same thing to a daily sports newsletter’s fiftieth issue. The mockery is the point. So is the work.”
— The Sports Page, Semicentennial EditionOver-Reactions and Under-Reactions
Over-Reactions
Issue #49 (Mets-Braves asymmetry), which ran yesterday, framed the Mets’ rivalry-side baseball as a long, quiet act of being out-organized. That framing was correct for the historical sample. It is, however, a strange piece to publish in a week where the Mets just swept Detroit. A reader who skipped Issue #49 and watched the Tigers series would have thought the team was on the rise. A reader who read it would have read about a quarter-century of finishing second to Atlanta. Both readers, by Friday night, were watching a 111-mph comebacker fracture the rotation. The piece is right and the timing was odd; the data resolved both.
Under-Reactions
The Mets’ injury accumulation, before Friday, was the most under-covered story in this newsletter’s May. Senga (back), Lindor (calf), Polanco (Achilles/wrist), Robert (back), Alvarez. Five names, five reasons the team Issue #38’s log5 model projected does not actually exist anymore. Each individual injury looked manageable. The aggregate was a depth crisis that the model could not encode and that we did not write about. Then Holmes fractured a fibula on Friday and the under-reaction became unmissable. A methods note for the next 50 issues: when injuries pile up, write about the pile, not just each piece.
What Readers Read · May 10–May 16
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 Almanac · A Milestone Observance
Every fiftieth issue of this newsletter is, by our own coinage, an “edition” in the same fake-Latin manner that the federal government has decreed 2026 the semiquincentennial of the Republic. This is The Semicentennial Edition. In observance, the Almanac looks back at what was actually happening at the five prior fifty-year marks of our national life.
- 1776(250 yrs ago) — The Continental Congress sat in Philadelphia, debating whether thirteen colonies could become anything at all. The vote that mattered came on July 2.
- 1826(200 yrs ago) — John Adams and Thomas Jefferson, the last surviving signers of the Declaration, both died on July 4 — fifty years to the day after independence. Adams’s last words were reported to be “Jefferson still survives.” Jefferson had died five hours earlier.
- 1876(150 yrs ago) — The presidential election that fall was decided not at the ballot box but by a fifteen-member electoral commission. Samuel Tilden (D) won the popular vote. Rutherford B. Hayes (R) was awarded the White House in exchange for ending Reconstruction.
- 1926(100 yrs ago) — Babe Ruth hit 47 home runs for the Yankees. Lou Gehrig hit 16. Together they remade what an offense could be.
- 1976(50 yrs ago) — Jimmy Carter, a peanut farmer from Plains, Georgia who had been polling at 1% a year earlier, accepted the Democratic nomination in New York and went on to beat an incumbent.
The Road Ahead · Issues #51 Through #100
The next fifty issues will take us from May 17 through roughly the first week of July. That window contains: the back half of the MLB regular season’s most-informative stretch (the stabilization point for batting averages, late June; for ERA, end of June), the NFL OTA period and the post-draft talent shakeouts, the NHL Stanley Cup Final, the Belmont Stakes, and a Notre Dame football preseason that will, by July 1, be the dominant editorial story for the rest of the calendar year. The newsletter expects to cover all of it at roughly the same daily pace.
The next milestone edition lands at Issue #100, the Centennial — on a Tuesday or Wednesday in early July, depending on holidays and Sunday Edition arithmetic. Every fiftieth issue from there forward will get the same treatment: a silly Latin label, an Almanac block with anniversary years, and a methods retrospective on whatever has accumulated by then. The federal government’s actual Semiquincentennial Edition for the country lands on Issue #250 — sometime around Christmas 2026 — and the joke will, on that day, be at its full strength.
Until then: small samples lie. Ratios are dangerous. Stabilization takes longer than fans want. Bayesian frames are useful. Folk theories aggregate badly. We have been wrong, and we will be wrong again. The work continues.
“Fifty issues down. Four hundred fifty to go. See you Monday.”
— The Sports Page, on the long game and the round number
