Our model projects the 2026 Yankees at 88 wins and the Orioles at 81. FanGraphs has the Yankees at 87 and the Orioles at 84. The two systems mostly agree at the top of the division — but they disagree meaningfully on Baltimore. If you've looked at both sets of projections and wondered who's right, the honest answer is that they're both right. They're just answering different questions.
The Core Difference
Our ELO system is backward-looking. It's a rolling summary of how teams have actually performed, updated after every game and carried forward season to season. A team's 2026 projection is anchored to their 2025 results, modified by offseason roster changes and — once the season starts — starting pitcher matchups and game-by-game outcomes.
FanGraphs projections (ZiPS and Steamer) are forward-looking. They don't care much about a team's 2025 win-loss record. Instead, they project each individual player's stat line based on historical performance, aging curves, and regression to the mean, then aggregate those projections into team-level win totals.
One system asks: how good has this team been? The other asks: how good should this roster be? Those are meaningfully different questions, and the gap between the answers is where the interesting analysis lives.
Where the AL East Stands
Here's how the end-of-2025 ELO ratings — shaped by actual game results — feed into our 2026 projections, alongside FanGraphs' Depth Charts projections for comparison:
Team | End-of-2025 ELO | ELO Projected Wins | FanGraphs Projected Wins | Playoff Odds |
|---|---|---|---|---|
NYY | 1675.6 | 88 | 87 | 74.4% |
TOR | 1615.4 | 86 | 86 | 60.7% |
BOS | 1622.4 | 85 | 86 | 63.1% |
BAL | 1510.3 | 81 | 84 | 37.1% |
TB | 1493.9 | 78 | 81 | 20.1% |
The two systems are remarkably close at the top — within one win for New York, Toronto, and Boston. The real disagreement is at the bottom: our model has Baltimore three wins lower than FanGraphs, and Tampa Bay three wins lower. That pattern isn't a coincidence, and understanding why it exists is the point of this article.
The raw 2025 season results left a 165-point ELO gap between the Yankees and Orioles. But those raw ratings aren't what drives the projections you see on the site. Before Opening Day, the model constructs a preseason ELO for each team by blending two independent signals in a 50/50 split:
Signal 1: Mean-reverted ELO. The model regresses every team's end-of-season rating 40% toward the league average of 1500. This reflects the historical tendency for teams to regress toward the middle — last year's best team rarely stays the best, and last year's worst team rarely stays the worst.
Signal 2: FanGraphs projected WAR. We pull FanGraphs' Depth Charts projected WAR for all ~1,400 MLB players, aggregate by team, and convert each team's total WAR to an ELO-scale rating centered on 1500. This is where forward-looking talent information enters the model — aging curves, injury risk, and expected playing time are all baked into the FanGraphs projections we consume.
The blend is deliberate. Mean-reverted ELO carries information about organizational quality, coaching, and intangibles that WAR doesn't capture. WAR captures roster talent that end-of-season ELO hasn't yet processed — particularly offseason moves. Neither signal is sufficient on its own; the blend outperforms either individually in backtesting.
This preseason layer compresses the Yankees-Orioles gap significantly, but it doesn't eliminate it. The result is a 7-win projected gap between the Yankees and Orioles in our model — compared to just 3 wins in FanGraphs' projections. That difference is where the interesting divergence lives.
The Orioles: Why the Models Disagree Most
Baltimore is where the philosophical gap between these two systems shows up most clearly.
The Orioles finished 2025 with a disappointing record despite a talented young roster. Our model sees those results and responds accordingly — ELO dropped to 1510, barely above the league average of 1500. The preseason blend gives partial credit for offseason acquisitions, but the underwhelming 2025 season remains a dominant signal. The model projects 81 wins.
FanGraphs sees the same team differently. It doesn't penalize Baltimore for last year's record — it projects each player individually. Gunnar Henderson gets a 6.0 WAR projection regardless of the team's 2025 win total. Young arms get projected along their aging curves. The depth chart gets modeled position by position. The result is an 84-win projection — three wins higher than our model, and a meaningful gap in a tight division race.
Three wins might not sound like much, but in a division where FanGraphs has the top four teams separated by just three games, it's the difference between a fringe contender and a legitimate playoff team. Our model sees an 81-win team sitting seven games back of the Yankees. FanGraphs sees an 84-win team sitting three games back. Same roster, very different positioning.
The gap comes down to a fundamental question: does last year's team record carry meaningful information beyond the talent on the roster?
Our model says yes — there are intangibles that team-level results capture (clubhouse dynamics, managerial strategy, organizational execution) that individual player projections miss. FanGraphs says no — or at least, not enough to override what the talent says.
Factor | Our Model | FanGraphs |
|---|---|---|
Unit of analysis | Team | Individual player |
Primary signal | 2025 W-L record | Player skill projections |
Luck adjustment | None — outcomes are the signal | Regression to the mean |
Aging and development | Not captured | Explicit aging curves |
Injury recovery | Not captured | Playing time projections |
Internal prospects | Not captured | Projected from minor league stats |
Roster depth | Not captured | Full depth chart modeled |
Offseason moves | WAR-based ELO shift | Player-level additions and subtractions |
For a team like the Orioles — young, talented, coming off a season that likely undershot their true talent level — FanGraphs is structurally more optimistic. Three wins more optimistic, to be exact. Our model needs to see the wins before it gives Baltimore full credit.
The Blue Jays: A World Series Team Still Chasing the Yankees
Toronto made the 2025 World Series. Our model projects them second in the AL East at 86 wins — still two games behind the Yankees. How can a team that just played in October's biggest stage not be the division favorite?
The answer lies in ELO's multi-year memory. A single season — no matter how good — can't fully erase the prior baseline.
Team | End of 2024 ELO | 2025 Change | End of 2025 ELO |
|---|---|---|---|
NYY | 1553.4 | +122.2 | 1675.6 |
BOS | 1499.3 | +123.2 | 1622.4 |
TOR | 1466.1 | +149.3 | 1615.4 |
BAL | 1536.3 | -26.0 | 1510.3 |
TB | 1559.2 | -65.4 | 1493.9 |
Toronto's 2025 gain of +149.3 ELO points was the biggest jump in the division — the kind of surge you'd expect from a World Series run. But they started 2025 from a deep hole: an end-of-2024 ELO of just 1466.1, the lowest in the AL East after a disappointing 2024 season.
Meanwhile, the Yankees entered 2025 already at 1553 and compounded from there. Toronto climbed faster than anyone, but they were climbing from the bottom. The gap narrowed dramatically — from 87 points at the end of 2024 to 60 at the end of 2025 — but it didn't close. After the preseason blend, that remaining gap is why the model still projects the Yankees on top.
FanGraphs wouldn't carry that 2024 penalty forward. A player-level system gives full credit for the roster that made the World Series, projects each Blue Jay based on individual talent, and doesn't discount projections because of a season that's now two years old.
This is ELO's recency bias paradox: the system weights recent results heavily, but "recent" still means the cumulative history, not just the last season. A team needs consecutive strong seasons to reach the top of the ratings. One breakout year moves the needle dramatically — Toronto's 149-point gain proves that — but it doesn't erase history entirely.
The Yankees: Where Both Models Agree
The Yankees are the one team where ELO and FanGraphs see essentially the same thing. Our model projects 88 wins; FanGraphs projects 87. Both systems put New York at the top of the AL East.
That convergence is notable because the two systems are arriving at the same answer from very different directions. ELO sees a team that dominated the 2025 regular season, carrying a 1675.6 rating — the highest in the division by 60 points. The preseason blend compresses that advantage (which is why the model projects 88 wins, not 97), but it doesn't eliminate it.
FanGraphs sees reasons for that compression:
Aging stars projected to decline. Aging curves pull individual WAR projections downward for players on the wrong side of 30.
Bullpen regression. A historically good bullpen season is likely to regress toward the mean — something player-level projections account for explicitly.
Overperformance. Key players whose 2025 stats exceeded their true talent level get regressed back toward career norms.
The result is convergence from opposite directions: ELO starts high and gets pulled down by mean reversion, FanGraphs starts with individual talent and builds up. Both land at 87-88 wins. When two fundamentally different systems agree, it's a stronger signal than either one alone — the Yankees are probably a high-80s win team.
Which Is More Accurate?
Neither system is definitively better. They capture different information, and their strengths and weaknesses are almost perfectly complementary.
Where ELO has the edge:
Captures intangibles that are hard to quantify at the player level — clubhouse chemistry, managerial strategy, organizational execution
Self-correcting during the season as game results update ratings in real time
Simple and transparent: one number per team, fully automated, no subjective inputs
Where FanGraphs has the edge:
Forward-looking: projects what teams should do, not just what they did
Captures player-level nuance — aging, regression, injury recovery, prospect development
Less anchored to prior-year results, which makes preseason projections more responsive to roster construction
Where ELO falls short:
Can't distinguish luck from skill in team records
No player development or aging model — a 22-year-old and a 37-year-old are treated identically
Preseason projections are the noisiest point of the year
Where FanGraphs falls short:
Playing time projections are inherently uncertain — one injury reshuffles the entire depth chart
Doesn't capture team-level synergies or the kind of organizational momentum that shows up in win totals but not individual stat lines
Projection systems disagree with each other (ZiPS and Steamer often differ by several wins)
If you want a single takeaway: trust FanGraphs more in February, trust ELO more in July. Player-level projections carry more information before games are played. But once the season is underway, ELO's game-by-game updating is a powerful signal that incorporates everything happening on the field.
What This Means for the Season Ahead
Our model is at its weakest right now. In the preseason, the projections are built entirely from historical ELO and FanGraphs WAR forecasts — no 2026 game results have entered the picture yet.
But the preseason anchor doesn't persist all year. As games are played, the model gradually fades from the preseason ELO toward each team's in-season rating over the first 100 games:
Game 1: The effective rating is almost entirely the preseason ELO.
Game 50 (~late May): It's a 50/50 mix of preseason and in-season performance.
Game 100 (~early July): The preseason signal has fully faded. The model relies entirely on current performance.
The fade rate of 100 games was determined through backtesting across the 2024 and 2025 seasons. Here's what that means in practice:
April–May: ELO adjusts game by game, but the preseason anchor still carries significant weight. A hot start by Baltimore would begin closing the gap with the top of the division, but it won't override the preseason overnight.
By mid-season: The model reflects mostly current-year results. Teams that are genuinely better or worse than their preseason projection will have separated from the pack.
By September: Current-season performance is the entire signal. The preseason blend is gone, and the projections closely track reality.
The model will naturally converge toward reality as the season unfolds. If the Orioles are truly an 84-win team — or better — the ELO system will see it within a few months. If the Yankees are due for regression, the losses will show up in the ratings game by game. Right now the two systems disagree most on Baltimore and Tampa Bay; by July, the games will have settled that argument.
That's the tradeoff we've chosen: less speculation in exchange for more responsiveness. The projections you see today are a starting point. Check back in June and they'll tell a very different — and much more confident — story.
