Where the console came from
Supervisory Control and Data Acquisition did not begin as a philosophy of knowledge. It began as a practical problem for electric utilities and pipeline operators in the 1950s and 60s: a substation two hundred miles away, a valve station in open country, and a telephone line as the only channel back to the control room. Telephone reporting was too slow to act on. By the time a dispatcher heard that a breaker had tripped, the fault had already propagated. The industry needed to know plant state continuously, not periodically, because a system that stops being observed becomes a system nobody can control.
Modicon's programmable logic controller, 1969, replaced relay panels with reprogrammable logic that could report as well as switch. Honeywell's TDC 2000, 1975, put microprocessors on a shared data highway so a plant could be watched from one console instead of walked with a clipboard. Neither invention had anything to do with intelligence in the cognitive sense. Both were answers to a narrower question: how do you keep a moving process inside safe limits when you cannot stand next to it. The answer, arrived at independently across utilities, refineries and pipelines, was the same in every case. Poll everything, all the time, and treat every reading as provisional until the next one arrives.
The dugout has the same problem
A performance analyst working for a professional club faces a version of the same question, with a ball instead of a fluid. The process is a match, or a season, and it does not hold still. The analyst is expected to know, at any moment, what the opposition tends to do, who on their own squad can play ninety minutes, who is arriving in the transfer window and what a rival scout has been sent to watch. None of that is static. A left-back who overlapped relentlessly in September may have been told to stop in October. A striker's underlying numbers may look identical to August while his hamstring load readings tell a different story entirely.
The characteristic failure in this discipline is specific and recurring: a game plan is built on a tendency the opponent has already abandoned. It happens because the report that built the plan was accurate when written and treated as though it stayed accurate. The opposition's press trigger, filmed in September, gets coded into a dossier, briefed to the players on Thursday, and executed on Saturday against a team that changed its pressing rule three matches ago. The information was not wrong. It was unrenewed. That is a staleness failure, and it is structurally identical to a plant operator working from last month's pressure log while the transmitter reads something else in real time.
Three generations of scouting
Set the analogy out plainly, because the correspondence is exact enough to do real work.
| Generation | Sports analytics equivalent | What it cannot do |
|---|---|---|
| Large Language Model | The end-of-season report, or the dossier compiled before the window closed | Tell you what the opponent is doing this week |
| Large World Model | The live tracking feed during the ninety minutes in front of you | See anything outside this fixture, this half, this camera coverage |
| Large Universe Model | Tracking data, injury bulletins, transfer activity and opponent tendencies, all still running, each tagged with source and date, each decaying in confidence until refreshed | Nothing, structurally — but everything, practically, still has to be judged worth trusting |
A dossier compiled before deadline day is a plant historian exported to a file: an honest record as of a date, useless for tonight's setpiece routine. The live tracking feed during a match is the operator's console during a shift: rich, real, and bounded to the fixture on screen. What a serious analytics department actually tries to build — training-ground load sensors, medical bulletins, scouting reports on next month's opponent, market intelligence on who a rival club is bidding for, all held as beliefs with a timestamp and a confidence that decays until the next update — is the third architecture. It is not an aspiration borrowed from artificial intelligence research. It is the same thing SCADA became in industry, arrived at for the same reason: a process that will not hold still cannot be managed from a single sampling.
The pitch is not a refinery — and that is exactly why the streams multiplied
The obvious objection lands hardest here. A refinery has a finite, engineered tag list and known physics. A match has neither. Football is open-world: new formations, new signings, new injuries, referees who interpret the same law differently week to week. Generalising from forty thousand instrumented tags to an opposition's entire behavioural repertoire smuggles in a closure assumption that a football pitch will not honour.
The closure objection is correct about the refinery and wrong about the direction of travel. Refineries did not stay closed either. Tag counts went from hundreds to hundreds of thousands precisely because the engineered boundary kept failing to contain the actual causes of upsets — ambient temperature, feedstock assay, grid frequency, none of them inside the original plant boundary, all eventually pulled in because upsets kept originating outside the fence. Sports analytics departments followed the identical trajectory, on a shorter timescale. Video coding alone gave way to video plus physical tracking; physical tracking gave way to tracking plus medical load data, because non-contact injuries kept originating in training sessions the match footage never showed; medical data gave way to medical plus market intelligence, because a tactical plan built around a player who was about to be sold on deadline day was worthless the moment he left. The boundary of "what we need to watch" has moved outward every year of the last two decades, never inward. That is the leak the industrial history predicts, not a coincidence borrowed to flatter the analogy.
The alarm-flood problem, restated for a dugout
The second objection is the sharper one, and it should be conceded almost in full. Continuous observation has a well-documented failure mode: too many streams produce too many alerts, and operators drown rather than decide. Texaco's Milford Haven refinery, 1994 — two hundred and seventy-five alarms in eleven minutes before the explosion. The lesson generalised across control engineering afterwards was that raw intake volume is not the bottleneck. Judgement capacity is.
The same flood happens to analytics staff. A modern department can receive tracking data on every player at 25 frames a second, a medical bulletin every morning, transfer rumours daily, and a tendency report on next weekend's opponent updated after every one of their matches. Given all of it at once, an analyst under a Thursday deadline does exactly what an overloaded operator does: reaches for whatever was flagged loudest, not whatever matters most, and the plan gets built on the September tendency because the September report happened to be the one sitting finished in the folder while the October update was still half-coded.
Industry's answer to the alarm flood was never to switch sensors off. It was alarm rationalisation — deciding in advance which readings deserve interruption and which merely deserve logging, then suppressing or shelving the rest until context makes them relevant again. The sports-analytics equivalent is decay-weighted trust rather than raw recency: a tendency report loses standing the longer it sits unconfirmed by recent matches, an injury bulletin outranks a tactical note the moment a starter is doubtful, and nothing gets briefed to players without a date attached, so a coach can ask "when was this true" and get an answer. That is more processing sitting above the same intake, exactly the distinction the industrial case draws between having the streams and knowing what to do with them.
The floor, not the finished job
None of this claims that watching everything produces a winning gameplan. Milford Haven had abundant instrumentation and still exploded; a club can have every stream running and still lose to a tactic nobody flagged, because indication is not comprehension. What the recurrence across two very different practices does establish is narrower and harder to dismiss. Continuous, provenanced, multi-stream intake — tracking data, injury reports, transfer activity and opponent tendencies, each tagged with when it was true and how much to trust it, none of it ever treated as final — is the floor an analyst needs to stop building plans on tendencies that no longer exist. Getting from that floor to a correct read on Saturday's opponent is still the hard, unfinished, human part of the job. The ladder has a top rung on the intake axis. It does not do the analyst's thinking for her.