Honest numbers.Here is where they stand.

What a computer-made clip is

We make a pitch on a computer, so we know its true speed exactly. The clip has a ball, a pitcher in front of it, a background, camera noise and iPhone-style video compression. Then VeloVision measures the clip and we compare its answer with the true one.

That is a good way to find mistakes fast, and to see which setups work. But it is easier than real life. Real clips add real light, real seams, a phone that may shake, and a distance you measured by hand. So these results are the best case. Real results can only be the same or worse.

What the test clips showed

All of these are on computer-made clips. They are not measurements of real pitches or swings.

Pitch speed

On 12 computer-made pitches filmed at 240 frames a second, 11 were measured and one was refused. The 11 were off by +0.15 mph on average, with a typical spread of about 0.8 mph and the largest miss 1.6 mph. In short: within about 1 to 1.5 mph, on computer-made clips.

Hitting (Experimental)

On 60 computer-made swings, 31 were measured. Exit speed was off by +1.3 mph on average (typical spread 1.2, largest 3.7). Launch angle was within 1.6 degrees. The rest were refused rather than guessed. Distance was off by about 4 feet on average (typical spread 6), before the extra uncertainty of wind and spin.

Spin (Experimental)

Spin is the hardest number. On computer-made clips with a clear, bright ball, the seam reader gave an answer for 15 of 16 pitches, with total spin about 4 percent low (typical spread 9 percent). It needs the ball to be big and clear for a dozen frames, and it cannot read a ball that is small or hidden.

What the ± means

The ± is the app’s own estimate of how far off a number could be. A speed of 75.3 mph ± 1.0 means the app expects the true speed to be about 74.3 to 76.3 mph. A bigger ± means it is less sure, for example when the ball was small, hidden for part of the flight, or you did not enter a distance.

• On the computer-made clips, the true speed stayed within twice the ± on all 11 measured pitches.
• That has not been checked on real pitches. Until it is, treat the ± as the app’s honest guess, not a guarantee.
• When the app cannot measure reliably, it shows No speed and says why, instead of guessing.

Why some numbers say Experimental

Spin, movement, release angles and hitting numbers are in the app, but they have only been run on computer-made clips. We label them Experimental until they have been checked against real pitches and swings.

Pitch speed is not labeled Experimental, because it is the most tested number. It has the same limit: it has not been checked on real pitches yet.

How real-world testing will work

We will film in the VeloVision app, the same way you do, and compare it with something we can trust.

• Speed: a radar gun sits behind the net, in line with the pitch, aimed back at the phone. Someone writes down its reading after every throw. We plan about 24 pitches, easy to full effort, fastballs, changeups and curveballs, and we compare the app’s speed with the radar’s.
• Setup: the phone is placed as in the filming guide. The distances are taped, not guessed.
• Spin: a baseball spun on a drill, with a flag on the shaft that turns once per turn of the ball, so we know the true rate. Also a ball marked with a stripe of tape, so we can count its turns frame by frame.
• Hitting: a radar behind the hitter for exit speed where it can see the ball, and a tape measure for where the ball lands. Launch angle can only be checked indirectly.

Some things cannot be checked this way. There is no home reference for how much a pitch breaks, so for movement we can show evidence (the same pitch type repeats, different types differ the right way), but not an accuracy number. And a radar gun has its own error, which we will say alongside the results. 20 swings is a first look, not an accuracy figure.

We have not set dates for this.

Results

Results from real pitches will be posted here.

If the real errors are bigger than the computer-made ones, we will say so, and change the ± and the Experimental labels to match.

Ask us a question