System stability

There’s an idea I’ve never been able to get my head around, called the ‘Lindy effect’, which (to crib from Wikipedia) “proposes the longer a period something has survived to exist or be used in the present, the longer its remaining life expectancy.” The idea is that if something has been around a long time, the end of it is usually a long way away.

And yet, all things (usually) come to an end.

So it’s strange to see a dominant global force, a force who’s been at the very top of the tree for a long time, floundering at the moment like Manchester City men’s team. Early diagnoses pointed to a single factor, Rodri’s injury, although there are, of course, surrounding factors. Maybe most system failures come down to one central factor, but the magnitude of their impact often seems to depend on how strong the rest of the system is.

Being in the tech side of football gives an interesting perspective on this: large (and/or experienced and conscientious) software companies use ‘chaos engineering’ to test their systems. (Microsoft Azure even has a product called Chaos Studio, which is how my colleagues refer to my presence in meetings). (That is a joke).

Of course, software is an easier type of system to test for failure than others. But you can see the value.

If nothing else, the chaos scenarios can force you to think through problems that you might unconsciously file in the box of ‘not going to happen’. What happens if your data provider suddenly changes the structure of their API; what happens if your Ballon d’Or nominated central midfielder ruptures their ACL.

These ‘what ifs’ might seem annoying, but many systems may seem stable not because of design but because of lack of testing. Rodri’s absence tests parts of the City system in ways, and with a frequency, that they simply hadn’t faced before.

This goes beyond the on-pitch action too. I’ve just finished reading Miguel Delaney’s recent book ‘States of Play: How sportswashing took over football’. In multiple cases, it seems like football’s governance structures were assumed to be reasonable until a moment in time where they were stretched further than it had been imagined they’d be stretched to. Regardless of your takes on those issues, it’s clear that systems (like around multi-club ownership) have been adapted to following new circumstances rather than adapting in anticipation of their potential arrival.

Obviously, the idea of scenario planning exists outside of software and football. But situations where chaos testing feels like it would’ve been helpful keep cropping up in the Premier League. Outside of Manchester City, there’s Manchester United who ended up with a marquee midfielder in Casemiro suddenly (though somewhat predictably) unable to cover the ground a central midfielder needs to cover. Then there’s Tottenham Hotspur, where an injury crisis has combined with a continued commitment to Ange Postecoglou’s pressing with fairly spectacular results.

Maybe this is where the biggest win for tracking data simulation could be. The idea, and practice, of using tracking data as a ‘prediction’ of player movements has been around a long time. Players and coaches don’t have a lot of time on the training ground. Perhaps, looking at their artfully constructed system on paper, the senate of Guardiola’s coaching staff thought that, yes, a combination of Bernardo Silva, Mateo Kovačić, and İlkay Gündoğan would able to cover for Rodri’s responsibilities. But maybe running that through tracking data simulations would have revealed vulnerabilities.

There are far more ‘chaos scenarios’ that you could run, too. Red cards springs to mind, but you could extend this to “what if our key player in X position is having a shocker” or “what if our press is horribly disjointed for no apparent reason”. “What if the front office sign 10 players and no-one knows each others’ movements”. ”What if our expensive new signing was actually a system player and also falls out with the manager”.

The idea of running simulations might seem fanciful to some (if so, why are you reading this, it’s the most Get Goalside idea going); but equally there’ll be some who’ll think that using it for chaos testing would be wasting it. This second group will be wondering whether you could use it to test different tactical set-ups before a match. After all, Formula One teams simulate different race strategies and conditions.

Personally, I would imagine that simulating football matches will be a lot less precise than simulating F1 races, and therefore the results are gonna be fuzzier. Do you trust the system to precisely determine the dynamics at play of using one first-team wide forward over the other? Maybe one day. But giving you the broad strokes of how the team might react to unlikely, but drastic, circumstances could be valuable without needing that precision.

Ultimately, as Manchester City show us, no system is perfect and no system’s robustness is set in stone. (Evolving circumstances, like recently-injured or ageing players, can be a slow boil of fragility). Each system will have trade-offs of goal maximisation and risk avoidance too. But systems shouldn’t necessarily be assumed to be stable just because they haven’t broken yet.