Methods & data

Data & dates

The built-in player database was refreshed on October 18, 2026. It contains 1,447 major league player records and 1,189 prospect records. This is a dated snapshot; it does not update automatically.

By default the tools use the “Offseason view (2027 on)” and value 2027 onward, because the 2026 regular season is over. Remaining 2026 production and salary are zero. The valuation clock still starts in 2026, so 2027 surplus is discounted one year.

• Current-year projections: Remaining-season fWAR and salary are zero after the regular season ends. Recorded 2026 statistics remain available as projection inputs.
• Future projections: public ZiPS. For the 50 players in FanGraphs’ 2026 Trade Value Series, 2027–31 use that series’ published ZiPS as-is, with no Dugout Value adjustment. For everyone else, 2027–28 start from published future ZiPS plus the in-season bridge described in the research note, and later seasons come from an aging curve anchored to ZiPS.
• Contracts, roles, and availability: FanGraphs RosterResource, with model estimates where inputs are incomplete. Arbitration salaries are model estimates.
• Prospects: FanGraphs The Board and selected SoxProspects grades. Four players absent from the current Board retain their July 31 grades, with source dates and notes in their editors. Their organizations were checked against MLB transactions.
• Trade finder: roster needs and projected 2027 production. The offseason pool includes every other club; completed-season playoff odds do not identify sellers.

Download the dated model and data archive to reproduce the October 17 model with its July 31 inputs.

Reviewed option paths are preserved across refreshes. Missing roster records require transaction evidence or remain unresolved. Individual player editors show projection and contract sources. A database refresh date does not imply that every upstream source was published on that date.

Providers: FanGraphs projections, RosterResource, The Board, and SoxProspects.

Model assumptions

The price of a win. For major leaguers, the model converts each projected season of fWAR to dollars, subtracts salary, and adds up the seasons of club control. The default price is $12 million a win, 30% higher for each net win past two, and it rises 3% a year. A roster-spot allowance of 0.5 wins for regulars and 0.2 for relievers comes off each forecast first. These are assumptions, not observed trade-market prices.

That curve is richer than FanGraphs’. FanGraphs prices a win at roughly $7 million for the first win, $8.5 million for the second and $13 million after that. At default settings, before the 3% yearly rise, a 4-win season is priced about 14% above FanGraphs’ 2026 graduated scale ($47.4 million against $41.5 million) and a 6-win season about 16% above it ($78.6 million against $67.5 million).

The discount. By default, future surplus is discounted 10% a year while the price of a win rises 3% a year, matching the rates FanGraphs uses for its prospect values. A no-discount scenario is available in the model settings.

The July 31 comparison below shows how the defaults changed the board. Earlier versions of this site didn’t discount, which put Kevin McGonigle at $594.3 million. The discount alone brings him to $402.4 million, and dropping the rest of 2026 brings him to $382.4 million. It moves Aaron Judge the other way, because his costly seasons come last.

The reason for the change is that the two halves of the model were running on different clocks. FanGraphs’ prospect table is already discounted at 10%, but major league seasons weren’t, so a player’s value jumped when he graduated. Prospect values are now not discounted a second time. The discount clock starts at 2026, so 2027 is already divided by 1.10. The refresh updates player inputs without moving this annual discount clock. Prospect values still use the published July 2026 table; a new input date does not rebase that table.

Offseason view. The tools and trade finder value 2027 onward. Players with no modeled control after 2026 are excluded from rankings and suggested offers. Historical deadline calculations remain in the downloadable July-input archive. The deadline adjustment is disabled in the current tools because no regular-season 2026 production remains.

Seasons the projections don’t reach. Past the last published ZiPS season, value comes from the site’s own aging curve. That season is 2031 for the 50 Trade Value Series players and mostly 2028 for everyone else. In the July 31 comparison, $111.1 million of McGonigle’s $382.4 million, 29.0%, came from 2032–34. The tools mark those seasons.

Prospects. Prospect values use FanGraphs’ July 2026 surplus-value table by Future Value grade (Clemens, “Introducing an Updated Method for Prospect Valuation”), with small adjustments for Board rank, roster status and arrival. That table prices wins on FanGraphs’ graduated scale, which differs from the curve used for major leaguers. The common dollar scale is useful for comparison, but its accuracy across player types has not been tested against trades.

Ranges. Value ranges are rules of thumb, not probabilities. Each is the midpoint plus or minus an allowance that nobody has checked against outcomes. Range width depends on role, injury status and how far out the surplus sits. It does not yet depend on age, major league experience or projection sample size, so ranges for recent call-ups are too narrow. In the July 31 comparison, rookie McGonigle received ±20.8%, against ±17.0% for Julio Rodríguez. The difference is distance, not age. Injury labels widen a range without cutting its midpoint, and a no-trade clause flags an approval requirement without changing surplus value.

How to read the ranges →

See the full calculation and adjustable settings →

Where the model disagrees with FanGraphs’ Trade Value list

This comparison uses the archived July 31 player inputs. FanGraphs’ 2026 Trade Value Series, by Ben Clemens, ranks 50 players. The database carries each player’s rank: 48 major leaguers and two prospects, Jesús Made and Kade Anderson. Ranking those 50 by the model’s surplus value at default settings gives a Spearman correlation of 0.603 with the FanGraphs order. A correlation of 1 would be the same order. It was 0.600 before the discount and the offseason view.

That is a benchmark comparison, not a backtest. FanGraphs’ list is another model’s judgment, not what happened to anyone, and the database holds ranks, not FanGraphs’ dollar figures, so I can compare orders but not amounts. Agreement is no proof the model is right, and disagreement is no proof it’s wrong. The six largest gaps follow.

To find each driver I re-ranked the same 50 players four ways: with every salary set to zero, with no star premium, counting only seasons through 2031, and with no discount. The driver is the test that moves a player’s rank furthest toward FanGraphs’, and only if it moves him three places or more. The next two gaps are Brice Turang (FanGraphs 19th, model 42nd) and Yordan Alvarez (20th, 38th).

Four of the six fit one pattern: the model charges harder for salary (Yamamoto, Judge) and credits long control more (Basallo, Anthony) than FanGraphs’ order does. That is what surplus value means, and it is also the part of the model nobody has tested against trades.

What’s been tested

The repository contains a hitter projection audit last run July 21, 2026. A ridge correction failed the stated same-season uncertainty gate and performed worse than future ZiPS on the following-season holdout. It was not deployed. The smaller role-aware adjustment remains a conservative modeling choice. Its measured advantage had an interval that included no improvement. This page reports the audit as it ran on July 21. It hasn’t been rerun since.

Three other pieces of research are published. None of them feeds the trade values.

• A hitter’s cold zone isn’t a pitching plan: in 2024–25, opponents’ hitter-specific plans lined up with the model at a slope of 0.464 (95% interval 0.391 to 0.537), against 0.027 to 0.098 for shuffled labels. The test of whether that cost hitters anything came back null.
• Pitch shape knows something a first half doesn’t: public pitch traits improved second-half K-BB% forecasts by 2.6% in 2024–25 (95% interval 0.4% to 4.8%). The model’s RMSE was 4.6% higher than ZiPS in the 2025 benchmark.
• Three Statcast ideas that didn’t improve the forecast: Damage Conversion, Arsenal Synergy and spray-adjusted xwOBA, all null or worse.

Nothing here establishes pitcher accuracy, trade-price accuracy, or calibrated value ranges. The FanGraphs comparison above is a benchmark, not a test of any of them.

What’s missing

The tool is a decision aid, not a claim about a club’s private model. It does not fully model team payroll, competitive balance tax, option-year eligibility, service time and 10-and-5 rights, 40-man roster spots, qualifying offers, medical information, or each club’s preferences. Arbitration salaries are model estimates, and the error of that estimate has not been published. Its prospect roster adjustments are rough costs, not a complete 40-man roster simulator.

Contract paths with a player option or an opt-out, such as Julio Rodríguez’s and Juan Soto’s, are coded by hand. Refreshes now download to a separate review directory. Contract paths must be checked before those inputs replace the published data.

Before using a scenario outside this historical snapshot, verify transactions, injuries, service time, contract options, and trade protections. Editing an input creates a scenario; it does not refresh the underlying database.