We measure football, then we model it.

Z30 Labs builds an owned statistics warehouse, a Fantasy Premier League optimiser that targets overall rank rather than points, and a computer vision pipeline that turns broadcast footage into positions on a pitch, in metres.

Sign in to the app with GitHub. Invite-only while the data licensing is settled.

Figure 1 · Tracked positions, ground truth

0.0s of 12.0s

Twenty two players, a referee and the ball, on a 105 by 68 metre grid. This is real output from vision.synthetic, the simulator our calibration is measured against, not an animation drawn to look like one. The registration step recovers these positions from the rendered frames to a mean error of 0.17 m.

01 · Practice

Three things, one database

The three lines of work are the same project seen from different ends. The warehouse is why the models can be checked; the models are why the warehouse has to be right.

Sports statistics

Eleven seasons of player-match rows, club metrics and transfer history, with one crosswalk of club and player identity across every source. Every row carries where it came from and when it arrived, so a number can always be traced back to its origin.

Predictive modelling

Minutes, goal and assist rates, clean sheets, bonus and discipline, fitted per player and combined into a points distribution rather than a single figure. The spread is the part that matters, and it is reported alongside the projection.

Computer vision

Pitch registration from broadcast frames: recover the camera, project the markings, and put every tracked player on a coordinate grid in metres. Accuracy is quoted in metres on the pitch, never in pixels on a screen.

02 · Products

What you can use today

Invite-only while the data licensing is settled. There is no free tier, because a free tier would mean strangers reading data whose redistribution rights are not yet established, and that is the one mistake that is expensive to undo.

Fantasy Premier League optimiser

A multi-period integer program over transfers, the bench, the captaincy and the chips. It knows the budget, the three-per-club rule and what a hit actually costs over the horizon you care about. It optimises overall rank, which means simulating where a decision leaves you against the field rather than counting points.

Not on sale yet

Sign in to the app

Statistics warehouse

The same database the optimiser reads, with filters, a query console and integrity checks that run on every build. Eleven seasons of the Premier League, with the Saudi Pro League being added.

Bundled with the app

Open the warehouse

Vision pipeline

Registration, detection, tracking and per-player metrics in metres. Working end to end against simulated footage and measured against ground truth. Not yet pointed at broadcast video.

In development

Data licensing

Access to the warehouse itself, rather than to the product built on it. Blocked on the same licensing question that keeps the front door shut, and it will open in that order.

Not yet offered

03 · Research

Method notes and accuracy reviews

Short, and honest about it: the index is new. What is listed below is measured and reproducible from the repository. Where a result exists but the writeup does not, it says so rather than staying quiet.

DateTitleStatus
2026-09 Pitch registration accuracy on simulated broadcast footage Chamfer registration against a distance transform of the markings, scored in metres on the ground plane. Mean position error 0.17 m end to end, through render, detect, track, register and back-project. Measured
2026-09 Warehouse integrity: seven checks over 254,119 player-matches No duplicate player-matches, no orphaned keys, values within bounds, club minutes reconciled per match, lineage on every row, and 50,000 sampled point totals recomputed exactly from their own components. Measured
Optimising rank rather than points: a season-long backtest Walk-forward over eleven seasons against a hold benchmark, with chips played. The harness exists and runs; the writeup does not yet. Writeup pending
Where the model was wrong last gameweek A standing post-hoc review, published whether or not it flatters the model. Starts once the season is under way. Not started

04 · Contact

Get in touch

For access, licensing or anything else: zancoholdings@gmail.com