Frequently asked questions

No. A phone propped somewhere stable is enough. The one hard requirement is that it does not move during the recording — the court is found by comparing frames from across the whole match.

High and behind the back wall, angled down the length of the court — a balcony rail is ideal. From there the players rarely block each other and their feet stay visible. Get the whole floor, including both back corners, inside the frame.

The filming guide has diagrams.

30 frames per second is enough for positioning, rallies, heatmaps, T control, distance and work rate. 60 fps or higher also unlocks ball speed and shot classification. Most phones record 30 fps unless you change the setting.

That is how the analysis separates one game from the next: a long pause with nobody on the court. If both players stay on through the break, a whole evening can be read as a single game, and the per-game breakdown cannot be recovered afterwards.

The story of the match first: the result, how each game’s rallies went, the run that decided it, one line on what happened and one thing to work on. Then where points were won, who controlled the court, and whether it held up to the end. Effort figures and heatmaps are there too, for each player, per game and overall.

Every report also says how much of the footage was play, for how many rallies a winner could be worked out, and how much of the court was in frame. How each figure is measured is explained below.

Roughly two to five times the length of the recording. A 20-minute match is under an hour; a two-hour club evening is most of an afternoon. You are emailed when it is ready, so you do not need to wait on the page.

Only from 60 fps footage or faster. At 30 fps the ball is motion-blurred to the point of not being in the image, so those metrics are not offered rather than estimated badly.

Not currently. The report is charts and figures. A marked-up replay is on the list but not built.

From the serve. In squash the winner of a rally serves the next one, so the next server is the winner. The same player serving again from the same box means the point was replayed, so nobody won it.

No winner is given when a rally could have been missed in between, or the server could not be seen. That leaves about three in four rallies with a winner. It is not the official score, and the report says “from the serve” wherever it relies on it.

The share of rallies you won when you were serving, and when your opponent was.

Your match’s own rallies, split into thirds by length. “Long” means long for that match, so a fast pair and a slow pair are both compared fairly. Every game uses the same split as the whole match.

The player nearer the T as a rally finished, and how often that player won it. In the matches we checked, the player nearer the T won about two rallies in three.

The share of rally time a player spent more than half a metre deeper in the court than their opponent. Being pushed back usually means the other player is dictating.

The share of rally time a player spent within 1.5 metres of the T.

Rallies won and T control, game by game. For a single game it compares the first half with the second; for a recording whose games could not be separated, the first third with the last.

When only a few rallies count towards a figure, one rally moves a percentage a long way, so the count is shown instead. With too few, the figure says so rather than showing a number.

From the figures on the report, by fixed rules. They are not written by AI, so the same match always reads the same way, and every line can be traced to a number on the page.

Distance, pace and how quickly a player gets back to the T, all measured during rallies only. Useful to know, but they say how hard a match was, not who controlled it, so the report keeps them folded away.

It has been proven on several courts in different buildings, all filmed from behind the back wall. A glass back wall or a very different camera angle may not work yet — if the court cannot be found with confidence, the analysis stops and tells you, rather than producing numbers that are quietly in the wrong place.

Almost always because the court could not be found — usually the camera moved, or too much of the floor was out of frame. It is a deliberate refusal: a wrong calibration would make every number look reasonable while being wrong. Re-film with the whole floor in shot from a still camera.

The players did not leave the court between games, so there is no break for the analysis to split on. Step off court, even briefly, between games.

Yes — two tracked players. Doubles is out of scope.

MP4, QuickTime (.mov) and Matroska (.mkv), including the 10-bit HDR footage most phones produce. Very long recordings have an upper limit shown on the upload screen.

It is deleted once analysis finishes. What is kept is the analysis — where players moved, the rallies, the statistics — which contains no footage of anyone.

By uploading, you confirm everyone in the recording has agreed to it being analysed. An opponent’s name is never shown to anyone else unless they hold an account and have agreed to it. The video not being kept is the main protection.

Yes. A club or coach account holds many players; an independent player has a workspace of their own.

Back to the T is in early access while the analysis is proven on more courts, and pricing is not set yet. Request access and you get a workspace to try it in.