Muse, tell me about a complicated metric

Hello. I scrapped a couple of drafts recently and have just put abridged versions into a grab-bag here. Have fun.

1) Momentum and Compactness

It's been interesting during the World Cup seeing the interest around the momentum charts on the FIFA broadcast, particularly the different varieties of 'momentum' that people have uncovered while looking into them. FIFA's is here but imo final third passes serves as a decent, easy proxy.

Part of why I think that is because I, personally, don't like most momentum charts, tbqh. The degree of spikiness doesn't represent what I feel watching a game -- never has, and I built in variable averaging windows in Twenty3's version of the vis to accommodate this if desired -- but each to their own. Catalina Bush has pipped me to the post with a newsletter if you're interested in more momentum.

I also wonder about the difference in how a fan and coach may perceive, or simply what they'd desire, out of 'momentum' data. As a fan, my gut feeling is that momentum shifts about five times a match tops; as a coach, you're going to be looking for possible intervention points. That also means that as a fan you tend to be looking for a representation of what has happened or what is happening, while a coach is likely looking for a leading indicator they can double down on or avert.

Meanwhile, the World Cup also reminded me of my semantic disagreement with 'compactness' as a term. England's last half-hour against Argentina was very compact. Argentina were, in turn, often very compact against France. In both cases, the team were also not helping their chances of winning the match. A huddle of players is the ultimate form of compactness, save for rare Paul Tierneys, but you wouldn't set a defence up that way.

(This post would've used Emily Wilson's Odyssey translation as an analogy, of how translation is not a fixed science and how linguistic meaning can evolve)

2) DeepMind and Databricks

Google DeepMind have some kind of partnership with Palmeiras on their TacticAI. Databricks published a content marketing blog about 'Coach's Corner' apps built on their software (which share a passing similarity to the hobby app from this parish a couple of years ago). They feel like the most sophisticated (public) manoeuvres of Big Tech in football as something more than just a service provider.

As I've written about before, football data gets itself frozen in stalemates between data providers who need reliable income for costly collection, football clubs who want to hop around to whatever works best, and third-party companies who want to build stuff but need to integrate with data providers to do it.

At some point, the walls will come tumbling down. There were a number of ways it could've happened, but big Big Tech service providers offering football-native services is one of them. If Big Tech throws some of their excess capacity at the engineering (and sometimes clout) challenges of integrating with data providers, and that could free a whole lot of things up.

(This post would've used the siege of Troy and Trojan Horse as an analogy)

3) Requesting methodology in a vertical video age

Analytics is on Instagram (and YouTube). The 'From First Principles' podcast had a clip on momentum charts go fairly viral midway through the World Cup. Katherine Gilligan (synsation_) did a series on World Cup tech, adequately focusing on the footballs that have chips in (footballs have chips in now, it's so cool). Mariana Antaya (mar_antaya) did some reels making ML models to predict winners of some matches.

One interesting question this raises is how to link to methodology or sources on platforms that don't make links easy. This isn't really a new problem, because Twitter/X started suppressing external links shortly after its takeover, and LinkedIn appears to do something similar. But it is worse for people not to be able to easily follow links than it is if they can.

That said, it's nice that this kind of interest is finding a home.

On a related video theme, Catapult have launched an Impect video product. Another set of video to ask for methodology from.

(This post would've been wrapped in an analogy about authorial viewpoint, and how Nolan's movie is driving sales of Odyssey translations)

4) Age of Ultron

Code generation AI has continued to improve. It wasn't that long ago that I was likening it to power tools that allow people to make chairs. Based on my LinkedIn feed, it seems like it's now at a level where it allows people to (mostly) re-do their kitchen.

There are some things you probably shouldn't try if you were re-doing your kitchen yourself, mainly around the cooker, I imagine, but y'know maybe you'd get away with it. I don't know what this means. Like the thing about data providers, the winners of this are probably clubs and organisations who already had a robust data engineering base in place. You can create an API with AI really easily now, but the seams of where everything interacts with databases and authentication systems and unexpected data provider quirks are where knowing how to spot bad work really pays off.

There's also a point made in a recent Pragmatic Engineer podcast episode. Dex Horthy, who is credited with coining the term 'context engineering', said this on model training:

"I don't think the models are smart enough -- I don't think we have the training and the benchmarking and the eval techniques -- to get models to write code that is more maintainable over time. [...] The problem with training models on maintainability is the cost function of bad architecture and program design can't be evaluated by running a unit tests, because it hits you three to six months later when you're like 'holy crap, this software has become so hard to change'."

Good architecture is important, but maybe there are things where good architecture is a priority to be managed.

(This post would have been wrapped in an analogy to the expertly-architected walls of Troy, undermined by bad human decision-making rather than architectural failure)

-

Go to the cinema,
Mark