About
Verdict Fencing makes professional-quality saber calls available to every fencer, from national team hopefuls to Tuesday-night club regulars, through the phone already in their pocket.
Verdict Fencing is built by William Spear, and it exists to give something back to a sport that gave him a great deal. Fencing paid for an education, built a career, and handed him most of the people he still knows. The club-night refereeing gap is the part of it he is in a position to fix.
He fenced saber. He was ranked No. 1 in the United States in junior men's saber through the autumn of 2011, made three junior and cadet world teams, and finished sixth at the 2010 Youth Olympic Games in Singapore, with a bronze in the mixed team event. He then spent four years on Columbia University's varsity squad, going 209–102 across four seasons and winning the Ivy League saber title in 2013. He qualified for four NCAA Championships, finished fifth in both 2012 and 2013, and took All-America honors three times. He co-captained the team in 2014 and 2015, and the 2015 squad won Columbia's NCAA team championship, the program's first since 1993; that team enters the Columbia Athletics Hall of Fame in October 2026. At senior level he held a US national ranking continuously from December 2014 until he retired in November 2019, peaking at seventh in the country. His strongest years were the junior ones, and he will tell you so.
After fencing he moved into software, and has worked since 2020 as an engineering consultant across defense, biomedical and scientific-computing projects. The relevant parts: he built a machine-learning system that detects, classifies and locates aircraft from audio captured by low-cost microphones; high-performance streaming simulators and distributed, fault-tolerant stream-processing systems for real-time data; full-stack graph-database applications serving medical researchers; and bioinformatics pipelines for assembling genomic sequence data at scale. Sensor data in, a classification out, at speed and under load. That is most of what an AI referee is.
That combination is the whole argument for Verdict Fencing existing. Plenty of people can build the computer-vision system; plenty of people know what a correct saber call looks like. Doing both is what makes it possible to be honest about which touches the model still gets wrong, and why.
Fencing has a referee shortage everywhere below the elite level. At club night, two lights go on, both fencers turn to whoever's nearby, and the point gets settled by seniority, volume, or a shrug. The sport's most interesting rule, right-of-way, is also its most argued about, and most fencers go years without reliable feedback on it.
Verdict Fencing is a computer-vision system that makes the call from a single phone camera. It tracks both fencers and both blades frame by frame, reads the scoring lights directly from the video, and judges the phrase the way a referee does, then explains its reasoning, so every touch teaches something. Saber first: it is the fastest weapon and the hardest right-of-way problem. Foil is next, épée after it.
It is not blade-trail visualization. The glowing blade-tip graphics on World Fencing League broadcasts draw where the blade went; Verdict Fencing decides who the point belongs to and names the action that won it.
As of August 2026, 31 fencers and referees, 7 of them international referees, have produced 27,906 completed labels across 23,752 saber touches, working since February 2026. Every label records the winner, the action, and the labeler's own confidence in the call.
Low-confidence calls go to a separate review panel: 5 validators, 4 of them international referees, have re-judged 3,258 of the hardest touches blind to the original call. Counting those reviews, 5,993 touches, about a quarter of the set, have been judged independently by two or more people. The rest carry a single confident label, and we would rather say so than imply blanket double-coverage we do not have.
Because a name on a call changes the call. Labelers work blind to each other, to the model's prediction, and to whose touch it is, so that seniority, reputation and deference stay out of the training data. In a sport where the referee's identity genuinely moves opinions, that blindness is the point.
The cost is the named-expert byline that normally signals credibility. So we publish the pool in aggregate instead: how many, what kind, how much, and how often two of them looked at the same touch. Those numbers come straight out of the annotation platform and are updated when they move.
Published numbers. An AI that makes officiating calls owes its users honesty about how often it is right. We publish our accuracy openly, per action type and including the calls we are still bad at, and the model says "not sure" instead of bluffing. When the numbers improve, we update the page; we never publish a projection as a result.
Verdict Fencing is in development, starting with saber. Early access opens to the waitlist first, free tier included. If you run a club and want to be part of early testing, we would especially like to hear from you at hello@verdictfencing.ai.
Fencers, coaches, and clubs. Early access opens this season.
Free tier at launch · One email when your spot opens — no spam.