Legend · Designed 2001
The gyroball
The pitch designed on a computer, mythologized into a monster, and quietly debunked by physics.
Signature pitchBullet spin, zero Magnus
The gyroball is real but heavily mythologized. It was not discovered on a mound but designed on a computer by a Japanese pitching coach and a computer scientist, who modeled its properties and published them in a 2001 book. The legend that grew around it, a "double-spin" pitch that breaks twice and that Daisuke Matsuzaka threw as a secret weapon, is the part that does not survive contact with the data.
A ball thrown with "bullet" spin, the spin axis aligned with the direction of flight, so the spin never passes over the face of the ball. With the angle between spin axis and trajectory at zero, the net Magnus force is essentially zero: the pitch falls on gravity and drag alone, with no lateral break. Its only honest real-world cousin is the accidental "back-up slider," which shares the same spin signature.
From Nathan and Baldwin’s analysis: "the net Magnus force on the ball is zero. The only forces acting on the gyroball are gravity and drag."
“I have done it in a game, but not too much. Sometimes accidentally.”
Daisuke Matsuzaka, asked whether he throws the gyroball; relayed from a 2006 Yahoo! Sports piece and cited in the SABR analysis.
The dramatic-break legend is false: with zero Magnus force there is no break to exaggerate. Two independent PITCHf/x studies of Matsuzaka’s 2007 pitches, one of 790 pitches and one of 586, both concluded he threw essentially no gyroballs. What it does do, against a hitter expecting a fastball, is drop roughly 15 inches more than that fastball would.
The two studies (SABR’s 790 pitches and Baseball Prospectus’s 586) reached the same conclusion from different samples. Both are kept here rather than merged.
Spin axis aligned with the trajectory, so no spin-induced force.
Alan Nathan elsewhere estimates closer to 18 inches.
A separate Baseball Prospectus study, same conclusion.
Filed the way every record here is: each figure season-stamped where it applies, confidence-labeled, and one click from its source. Where the reputation and the data disagree, the gap is shown, not smoothed over.