The Model

Projections

A Marcel-style baseline: three seasons weighted toward the most recent, regressed to the league, adjusted for age, expressed as wins per 600 plate appearances or 150 innings. We add translated Triple-A and Double-A performance, a regression target below league average, and extra regression for players with little MLB history. Where a player's results ran ahead of or behind his quality of contact (Statcast's expected wOBA), we move the projection part of the way back. Defense comes from outs above average and catcher framing.

How it has tested

Each number predicts 2026 using only earlier data. The flat guess gives everyone that season's actual average, which uses hindsight. We looked at 2026 several times while building the model, so these results are probably a little optimistic, and 2027 will be the first clean test.

Public stats still say little about players with little MLB history. Adding quality-of-contact data and Double-A lines more than doubled how well we rank those hitters, but a ranking score of .27 is still weak. That's where the numbers go quiet, and it's where a decade in Triple-A is supposed to earn its keep.

At the club level, pricing each player's actual playing time at his preseason projection tracks real records from 2022 to 2025: correlation .67, typical miss 9.7 wins, against 12.8 for guessing 81 for everyone. The newer inputs didn't improve this check: it moved within a few tenths of a win either way. That test knows who played, so it flatters the model, and it runs a bit narrow, understating how far good and bad clubs spread apart.

"Left on the table" has its own test. From 2018 to 2025, when the rule said a bench player was at least a win better at the plate than his club's regular, he out-hit that regular the next season 72% of the time, and the actual gap came in at 90% of the predicted one, so we cut the number by that much. That covers 200 pairs where both players still got real playing time, out of 338 flagged. It says nothing about defense or about players who never got a chance.

Injury model

From injured list placements and activations with major league clubs, we compute the share of each season a player lost, then resample real player-seasons by role and age, healthy seasons included. We leave out 2020, a short season, and 2021, which has far more placements than any other year for reasons we can't yet explain.

Season simulator

Each simulated season takes time away from players, then fills every job in order: who actually played there over the last two seasons (2026 counts double, so one injured year doesn't cost a regular his job), then the club's published chart, then replacement level, worth zero by definition. Published charts come second because in November they often list September call-ups ahead of regulars. Rest days go to the next man, and the best relievers get the highest-leverage innings.

Surplus value

Surplus is the market value of a club's projected wins, minus what it pays above the league minimum. Cost means luxury-tax value, which spreads bonuses across the contract and discounts deferred money. Each contract figure carries a source link in our data.

Without service time data, we sort players by years since debut. Early-career players get the league minimum. Arbitration salaries are estimates, set at a rising share of market value and checked against just five published projections, so they're rough. The price per win, $10.8M, is the median across 41 free agent deals, measured against our own projections, which run conservative. Some deals carry rules we haven't fully checked, such as the luxury-tax charge on front-loaded deals with opt-outs. Players we can't price are left out, which treats them as paid about market value. Treat surplus as a coarse read, not a precise valuation.

What's still wrong

• Rosters are a November snapshot: free agents are gone and Rule 5 additions haven't happened.
• Option years and service time aren't modeled yet, and they shape who can actually move.
• Most contracts are still unentered, and arbitration salaries are estimates.
• "Left on the table" weighs hitting only, and its test can't see players who never got a chance.

Sources

The information used here was obtained free of charge from and is copyrighted by Retrosheet. Interested parties may contact Retrosheet at "www.retrosheet.org". Player ID links come from the Chadwick Baseball Bureau register (Open Data Commons Attribution License). Rosters, transactions, minor league and 2026 statistics, and defensive measures come from MLB sources. We publish only our own derived results. Methods are on the Model page.

Model notes

Nov. 5, 2026. Added Double-A translations and an expected-stats adjustment. Both improved player projections in testing, most for thin-sample hitters, and left the club-level check about where it was.

Nov. 5, 2026. First version of projections, injury model and simulator.