aWAR

aWAR begins in 2016

Savant’s fielding leaderboard begins in 2016, so 2015 has no aWAR and no part of it.

Framing is 0 in 2016 and 2017

Savant’s framing leaderboard reads 0 for every catcher in 2016 and 2017, so framing adds nothing to a catcher or takes nothing from a pitcher in those seasons.

Framing shrank in 2026

In 2026, the first season of the automated strike zone, framing per pitch taken fell to 0.81 of its 2023 to 2025 level. Savant still frames a challenged pitch on the umpire’s original call.

Constants by season

Every season’s numbers, newest first. A finished season’s park factors and position values never change. The current season’s are rebuilt every night, so its aWAR moves with them.

Hitters

Pitchers

The ifFIP constant makes the league’s ifFIP equal its runs allowed per 9; on earned runs it would equal its ERA. The replacement shift is in wins per 9 innings.

Checks

The nightly job writes a season only when all ten pass; otherwise the last good version stays.

1. Hitters total 570 wins and pitchers 430, on both pitcher lines, times the season’s share of a 162-game schedule, within 0.05 wins. The table below adds them up as they stand.
2. Batting, park, double plays, baserunning, fielding, position and the league adjustment net to zero across hitters, within 0.1 runs.
3. Framing returned to pitchers equals framing credited to catchers, within 0.1 runs.
4. Our innings match Savant’s for every pitcher, or the difference is one of 5 known kinds, such as a reliever who records an out before his first pitch.
5. Savant’s fielding outs at each position are within 10 of our own count for every player on its board, and nobody missing from it has more than 30 outs. While a season is live, players who fielded on its last two game dates are left out.
6. Every player-season with a plate appearance or a batter faced has a row, no part is missing, and every range holds its aWAR.
7. Every club that hosted a game has park factors, and the league’s average run factor is within 0.005 of 1 once as many clubs have a factor as the season before.
8. Every position has a value each season, from at least 55,000 plate appearances.
9. Defense support charged to pitchers equals the range, arm and double-play runs Savant credits to fielders, within 0.1 runs.
10. The leverage table averages 1.000 over its plate appearances, and no state with fewer than 62 plate appearances stands alone.

Park factors

Runs and home runs per plate appearance in a club’s home games over its road games, pooled over the season and the two before it in the same ballpark, then pulled toward 1: three full seasons keep about 70% of the run gap and 72% of the home-run gap. Each plate appearance is adjusted at the park it was played in. Per plate appearance, not per game, because a hitters’ park’s extra plate appearances would otherwise count twice.

Runs

Home runs

Ballpark changes: ATH Oakland Coliseum 2016–24, Sutter Health Park 2025–26; ATL Turner Field 2016, Truist Park 2017–26; TB Tropicana Field 2016–24, George M. Steinbrenner Field 2025, Tropicana Field 2026; TEX Globe Life Park in Arlington 2016–19, Globe Life Field 2020–26; TOR Rogers Centre 2016–19, Sahlen Field 2020, split season 2021, Rogers Centre 2022–26.

Position values

aWAR uses FanGraphs’ position values. Ours, from five seasons of our own batting by position, swing more from one season to the next than sampling explains, the designated hitter’s most, so they are shown beside FanGraphs’ and not used.

Runs per 162 games. Ours assume the average player at every position is worth the same, and count pinch hitters with designated hitters.

Leverage

Built once from every regular-season plate appearance of 2016 to 2026, 1,905,138 in all, in 3,934 states of inning, half, outs, runners and the batting team’s lead, capped at four. A state’s leverage is its plate appearances’ average swing in win probability over the league’s, so 1 is average; it runs from 0.01 to 10.85. The 1,097 states with too few plate appearances share a value with their neighbours.

A reliever’s pLI is the average leverage of the batters he faced in relief, and his relief runs count at (1 + pLI) ÷ 2. Starters are unchanged.

The range

Where each part’s standard deviation comes from:

• Batting: the spread of his own plate appearances’ wOBA values.
• Double plays: binomial on his chances, times the run cost.
• Fielding, framing and baserunning: how well Savant’s values repeat from one season to the next, scaled to his playing time. Real changes in ability count as noise, so these parts err wide.
• Pitching: the spread of his strikeouts, walks and hit batters, home runs and infield popups, through the ifFIP weights. A reliever’s relief part is scaled by his leverage, and framing returned carries his share of his catchers’ framing noise.
• Park factors, position values, the leverage table and each season’s constants are treated as fixed.
• The runs-allowed line shows no range.

Checked for batting on 1,389 hitter-seasons of 500 PA or more: the gap between first-half and second-half batting runs, over its predicted standard deviation, has a spread of 1.055 (1 is exact), and 12.0% fall outside the 90% range. Pitching was checked the same way on 2025’s ifFIP, over 136 pitchers with 150 batters faced in each half: a spread of 1.118. The fielding, framing and baserunning parts, and hitters under 500 PA, are not calibrated.

Where aWAR differs from FanGraphs

Every place aWAR does something FanGraphs’ WAR does not, and why.

• Park factors are our own, from our pitch data, so a reader can rebuild them.
• Double plays are their own line. FanGraphs dropped them; the cost is small and the line is visible.
• Fielding and catcher framing are Savant’s fielding run value, our only fielding source. FanGraphs’ framing is its own model.
• Framing goes back to the pitchers pitch by pitch, by each one’s share of the catcher’s taken pitches, so a pitcher traded midseason carries the right share from each catcher.
• Reliever leverage comes from our own leverage table, built from our win probabilities.
• Every aWAR shows its 90% range; FanGraphs prints WAR without one.
• Pitchers also get a runs-allowed aWAR, the basis of Baseball-Reference’s WAR, with park and defense taken out per pitcher from the fielders behind him on each ball in play, where Baseball-Reference spreads one team figure across a staff.

Kept from FanGraphs: the split of 1,000 replacement wins into 570 for hitters and 430 for pitchers, which no source derives; the replacement levels of 0.12 wins per 9 innings for starters and 0.03 for relievers; half credit for reliever leverage; and leaving pitchers’ own hitting out.

What is missing

• A second fielding system, and any published error for Savant’s fielding: the fielding part of the range is our own estimate.
• Catcher blocking before 2018.
• A pitcher’s own fielding: a pitcher who also played the field has no hitter row, so his fielding runs reach no one (20 player-seasons from 2016 to 2021; the most, Michael Lorenzen in 2019, 2.5 runs).
• Catcher blocking and throwing on the runs-allowed line: defense support charges pitchers for the other seven fielders only.

What cannot be rebuilt

Savant’s fielding run value (range, arm, double plays, framing, blocking, throwing) and its baserunning run value are Savant’s own models. aWAR reads them as published, and no public data rebuilds them. Savant revises them, so a past season’s aWAR can move with no entry in the changelog below. Everything else rebuilds from public pitch data and the tables on this page.

Changelog

Each change to the method adds an entry naming every player whose aWAR moved by 0.5 or more.

• 2026-10-05: First version.