The objection that should win
Here is the case against continuous intake, put by a performance analyst who has shipped a hundred game plans and knows exactly how little of the data mattered.
Most of what decides a match is stable. A back four's shape, a coach's risk tolerance under a scoreline, a striker's preferred foot for the first touch — these do not change week to week. You do not need live tracking feeds and transfer-market chatter to know that a manager who has played back-three football for eleven years is not switching to a back four for one derby. A good model with strong priors absorbs a new opponent into an old category. Variety, properly counted, is not exploding. You are just counting too finely.
This is not a strawman. It is the working assumption behind a great deal of scouting practice, and it is often right. The claim of this page — that intake has a terminal form, and that sports analytics is a domain where the argument bites hard — has to survive that objection before it is worth anything.
Ashby's law, stated plainly
W. Ross Ashby, a British psychiatrist who became one of the founders of cybernetics, put the law in An Introduction to Cybernetics (1956). Variety is the number of distinguishable states a system can be in, counted on a logarithmic scale. If a disturbance can arrive in any of V(d) states and a regulator can produce V(r) distinct responses, the best that regulation can do is leave a residual disturbance of V(d) minus V(r). No amount of skill closes that gap. Only two things do: more distinguishable states in the regulator, or a filter that removes states from the disturbance before it arrives. Stafford Beer later took this into management theory, arguing that firms fail not from lack of effort but from a variety mismatch with their environment. A performance analyst is a regulator in exactly this sense. The disturbance is the opponent, and the residual is whatever the game plan gets wrong on matchday.
Where the objection is simply correct
Concede the physics first. A press-resistance metric built on pitch geometry and sprint speed does not need weekly refreshing; the geometry of a football pitch has not changed since Ashby wrote his book, and it is not going to. Biomechanical load thresholds, similarly, are close to stationary — a hamstring tears the way it tore in 1990. Where the disturbance source is physical, a model trained once can regulate for a long time, and continuous intake there is mostly wasted motion. This is the case for the coarse-grain reading: individuate finely enough and everything looks unstable; step back and much of sport looks like the same forces recurring.
But the decisions a performance analyst is actually paid for are rarely about physics. They are about what a specific opponent, with a specific squad, under a specific coach, is going to do in three weeks' time. That state space is not generated by mechanics. It is generated by ongoing human choices — recruitment, tactical experiments, injury-forced adaptation, a coach reading last season's failure and changing his mind. Ashby's law bites precisely where the disturbance is socially generated, because social systems manufacture new states on a schedule set by transfer windows and international breaks, not by any law that holds still for the analyst's convenience.
The tendency that expired
The characteristic failure of the trade is well known to anyone who has sat in a pre-match meeting: a plan is built to exploit an opponent's tendency to play out from the back under press, using footage and a data cut from two months prior, and the opponent has quietly abandoned that pattern after a bad result — new full-back, new instruction, or simply a manager's private loss of nerve. The plan is not wrong about anything that was true. It is wrong about elapsed time. The disturbance moved; the regulator's variety was fixed at the moment the data cut was taken, and the gap between them is the goal conceded in the second half.
The second objection: filtering instead of chasing
There is a more sophisticated defence of the frozen model, and it deserves its own hearing.
You don't need to track every stream in real time if you build a good enough interface. Standard opposition reports, templated scouting forms, a fixed pre-match checklist — these attenuate the variety before it ever reaches the analyst. That is what institutions are for. A club that has a disciplined, well-designed process can run on a slower data cycle than one flooded with raw feeds, and do it more cheaply.
This is correct, and it is not a small concession. Attenuation is real regulatory work, not a shortcut around it. A well-built scouting template that forces attention onto set-piece routines, out-of-possession triggers and substitution patterns strips a great deal of irrelevant variety before an analyst ever opens a laptop. Most of the value delivered by a sound analytics department is exactly this kind of filtering discipline, not raw data volume.
Two limits on it matter here. First, the filter is itself a regulator and ages the same way the plan does: a scouting template built around a back-three era will systematically under-sample the questions that matter once the league's tactical fashion shifts to inverted full-backs, because the template was never designed to ask about them. Second, opponents know the filter exists. A coaching staff that studies how it is being scouted will deliberately manufacture variation precisely where the standard template does not look — a new wide role introduced only in the final third, an out-ball pattern held back for a run of low-stakes fixtures and released for the one match it is meant to break. Attenuation narrows the problem. It does not remove the need for something, somewhere, to keep watching the parts of the game an opponent has an incentive to move.
What continuous intake actually buys, and what it does not
The corrective is not "watch everything, always." It is holding four streams open concurrently, each with its own decay rate, and reconciling them rather than trusting whichever arrived last.
| Stream | What it tells the analyst | Typical half-life of relevance |
|---|---|---|
| Tracking data (positional, per match) | Current shape, pressing triggers, sprint patterns | Days to a few weeks |
| Injury reports | Who is actually available to execute the plan | Days |
| Transfer activity | Personnel who change what the shape even means | Weeks to a transfer window |
| Opponent tendency logs across recent fixtures | Whether a pattern is a habit or a one-off | A few fixtures |
A game plan built on the tracking data alone, ignoring that the transfer window just removed the ball-playing centre-back the whole pressing trap depended on, has requisite variety in one channel and none in another. This is where the third technical objection lands, and it is the correct rebuke to any naive "more data" pitch: flooding an analyst with every available stream does not by itself produce a better plan if nothing downstream can hold four contradictory reads of the same fixture, date them, and decide which one governs on the day. Variety has to live in the response repertoire, not just arrive at the door. A department that adds feeds without adding the discipline of provenance — this reading is three days old, this one is from a fixture where the opponent rested six starters — can end up with more noise and the same blind spot.
The narrower claim
None of this proves that sports analytics needs infinite intake, or that a frozen model is useless in the sport. Set-piece geometry, load monitoring, biomechanical screening — these sit closer to the physics end and age slowly, exactly as the first objection argued. What Ashby's law licenses is narrower and harder to dismiss: wherever the disturbance is generated by an opponent's ongoing choices — recruitment, tactical revision, deliberate concealment of a new pattern until matchday — the regulator's variety must be replenished at something like the rate those choices are made, or the residual failure is not bad luck but arithmetic. Filtering helps, and should be built well. But the filter itself must be watched, and the streams behind it must stay open, because the alternative is a plan that is provably correct about a version of the opponent that no longer takes the pitch.