I’ll just explain, using Guidry’s 1978 season, which Andy and I were discussing, why WAR isn’t nearly as accurate or as valuable as measuring what the player actually did.
In 1978, Guidry’s record was 25-3, 22 games over .500. We don’t have to speculate about what a “replacement” player on the Yankees would have done. Jim Beattie was a replacement pitcher, a rookie who started 22 games, making the team in spring training after some injuries to veteran pitchers. He lost seven games in a row at one time and slipped out of the rotation for a while but ended up with the fourth-most starts on an injury plagued staff. Since Beattie actually played pretty much the whole season, he’s actually a notch above replacement (which WAR confirms by giving him a rating for the year of 0.7 WAR). But he’ll do.
Beattie was 6-9. So one way to measure his wins over replacement would just be Guidry’s wins (25) minus Beattie’s (6), or a WAR of 19. I think that’s a fair way, because a replacement player isn’t generally going to play a full year anyway. But that doesn’t count losses, and Beattie lost three times as many games in his 22 starts as Guidry lost in his 35 starts.
A better way to count in my view would be to compare how far they were above or below .500. The Yankees were 22 games over .500 with Guidry getting the decision, three games under when Beattie got the decision, so that way, Guidry’s WAR would be 25.
Or maybe you want to project Beattie’s performance over 35 starts, which would give him a record of 10-14. That would change the WAR to 15 just counting wins or 26 counting wins and losses.
But WAR doesn’t work that way. The WAR formula gives Guidry 9.6 wins above a mythical replacement pitcher. It tries to slice and dice a player’s contributions to the wins to figure hypothetically what that player’s number of wins is. And it’s flat wrong.