And it's live

The old phrase that 'history is written by the victors' has never been wholly true. The victors do get to promote their own versions and suppress (or destroy) those they don't like, but time and time again history ends up being told by folk who just keep telling it.

Anyway, I'll be giving the Research Stage's keynote talk at the Hudl Performance Insights conference in November, in no small part because of Get Goalside. If you're in the very small subset of people who are 1) in a realistic position to attend 2) have not already bought tickets 3) are meaningfully incentivised by that news, the link for the conference is here. Gaizka Mendieta will also be there on the day, if not in the audience.

One minor complication to blogging about this news is that I'd been fully intending to write a newsletter about the PySport Analytics Cup and the accompanying release of data from (non-Hudl company) Skillcorner, which includes real, honest-to-goodness, public football body pose data. But... ah, let's do it anyway.

I'll front-load the links:

Marc Lamberts was the first work I saw, demonstrating how the data can be used for scanning,

https://bsky.app/profile/lambertsmarc.bsky.social/post/3mvp2vi3mss2h

A post from MyGamePlan also demonstrated the idea of scanning, but also had a particularly interesting visualisation on player body orientation when they received a pass, fitting into 'open', 'closed', and 'half-open' categories.

And Gota Shirato took a slightly different approach with an exploration of the data, categorising how players turned when receiving the ball.

The data release is the coolest thing to happen in years. For my money, specifically six years: since Statsbomb dropped the Lionel Messi and Arsenal 2003-04 Invincibles data in their public datasets (in 2019 and 2020). Statsbomb, and other providers, have released other datasets in the intervening years, but these particular sets feel like a clear break from their respective prior eras.

Statsbomb (in its pre-Hudl days) started its open data in 2018, and it made a lot of stuff available. This was an era right in the midst of a lot of change, the mood music within and outside the analytics sphere changing rapidly (within: pitch control and possession value; outside: the success of Liverpool men's team).

But most of the publicly-available data (in GitHub and on websites) was still very 'current football'. The Messi and Invincibles datasets broke that mold, a little bit at least, by giving people the chance to compare modern day football to 'the past'.

It all meant that between 2014 and 2020, all this happened: expected goals was popularised online and then brought onto TV; models for quantifying on-pitch action with tracking data were showcased; and datasets giving a long-term view into football were published. There is a 'before' and there is an 'after'.

Skillcorner putting two matches of body pose data online feels similar, and is similarly wrapped up in other things.

The first is the PySport Analytics Cup itself -- an opportunity, like the Hudl Performance Insights Research Stage, to be provided with data and to present work to an audience of engaged professionals. Another is the set of tutorial notebooks that Skillcorner have provided, as well as the way AI can help with coding and general learning. Things happen when there are networks of ideas and support, and the facilitation of learning.

So, what could this data drop lead to? Assuming that the quality of the data is reasonable (which is a necessary 'if' to raise), there's a lot to delve into with biomechanics. Coincidentally, a football project coordinator at LALIGA recently posted (somewhat obliquely) about the organisation's use of skeletal tracking data in a project about ball-striking.

I'd be leaving a free square on the Get Goalside bingo card if I didn't mention defending, though. How do players look when they're involved in duels? Forget ball reception stance categories, what about tackle stance categories. Or, even better, will we finally get the answers to our 2019 proposals: 'who runs the weirdest' and 'which goalkeeper dives in the most needlessly showy way'?

I suspect that with the body pose data, more than ever, the approach to the work will be as, if not more, important than the work itself. On that, you could do a lot worse than looking at papers from the DTAI Sports Analytics Lab at KU Leuven for examples, including the recent paper 'How Long Do Actions Echo?'.  It's not an easy skill to gauge the scale of a project, in order to make sure that it's both meaningful and manageable, but it really helps your sanity if you manage it.

And a last item for increasing sanity, Catalina Bush's newsletter, The Post-Match Report is really good. Like if Get Goalside was more reliable [complimentary]. If you're subscribed to this, you should be subscribed to that.

See you in November, maybe.