communicoupling · concept 2 of 26
Why can't a simple rulebook govern a complicated situation? Ashby's law of requisite variety gives the hard answer: only variety can absorb variety. A controller can hold a system steady only if it has at least as many responses as the world has disturbances — and no amount of willpower substitutes for that count.
Picture the manager with three stock responses facing a hundred kinds of problem. The policy with two settings meeting a world of twenty states. The thermostat that can only turn on or off. We reach, reflexively, for the language of effort and attitude: they weren't trying hard enough, weren't flexible enough. Ashby showed the failure is arithmetic. The regulator was out-varied — outnumbered by the situations it had to meet, and no degree of resolve adds a response that isn't there.
This is why bureaucracies grow, why expertise looks from the outside like a suspiciously large repertoire, and why "keep it simple" has a hard floor beneath it. Below, set how varied the world is, how varied you are, and watch the ceiling on control snap into place — a ceiling you cannot climb over by playing more cleverly, only by matching the count.
Only variety absorbs variety
Ashby's law is easy to state and hard to escape. Call the number of distinct disturbances the world can present its variety, N. Call the number of distinct responses you can make M. If each disturbance needs its own answer — and in a world with no redundancy, it does — then a regulator with M responses can neutralise at most M of the N disturbances. The other N−M arrive, find no matching move, and pass straight through to the outcome. The best control any table can achieve is capped at M/N, and that cap does not care how you fill the table in.
Only variety can absorb variety. Not effort, not will, not attitude — those are ways of using a repertoire, never ways of enlarging its count. In the instrument, the doomed disturbances are the rows whose gold ring sits in a column you don't have. You can click every bright cell you like; their fix is literally off the board. When people say a rule is "too rigid" for a situation, this is usually what they mean, made exact: the rule has fewer settings than the situation has states, and the shortfall shows up as a fixed rate of failure no one can conscientious their way out of.
What to try
Drag M up one notch at a time and watch the ceiling climb in steps — each new response covers exactly one more disturbance. It reaches 100% only when M ≥ N. There is no smooth ramp and no shortcut; control arrives one unit of variety at a time.
With M < N, try to beat the ceiling by rearranging the table. Press Set the best table, then move cells around. The measured rate wobbles but never crosses the gold line. When you are out-varied, cleverness redistributes the misses; it cannot remove them.
Set M = N so full control is possible, then raise sensor noise. The ceiling falls below 100% even though your repertoire is complete. Variety in the sensing channel matters too: a response you can't aim is a response you don't really have.
Why institutions grow
Every new kind of disturbance the world invents demands a new kind of response. A tax code meets a new instrument; a clinic meets a new pathogen; a platform meets a new abuse. Each novelty raises N by one, and the law says that to hold control the regulator must raise M to match — a new form, a new role, a new exception, a new specialist who handles exactly that case and nothing else. The forms multiply not because someone loves paperwork but because each form is a response the world made necessary. Requisite variety is the honest name for a great deal of what we dismiss as bloat.
And the ratchet turns mostly one way. Disturbances are easy to add and hard to retire: the old scam still works, so the old countermeasure must stay, even as new ones pile on top. This is why mature institutions look baroque from outside and feel barely adequate from inside. Both views can be right at once. The repertoire is enormous and still out-varied, because the world it faces is enormous too. The question is whether the disturbance variety the rules answer to actually exists.
Attenuate or amplify
If control needs M ≥ N, there are exactly two honest ways to get there, and "just be more flexible" is neither. The first is to amplify your own variety: add responses, hire specialists, widen the repertoire until it matches. This is the expensive road, and it is what growing institutions actually do. The second, quieter and often wiser, is to attenuate the world's variety before it reaches you — standardise the inputs, filter out the cases you refuse to handle, buffer the shocks, force the hundred situations down into eight recognised types. Every intake form, every "please select from the following options," every triage protocol is an attenuator: it shrinks N so that a modest M can cover it.
Seen this way, simplification upstream is control, not the opposite of it. The regulator who standardises the world is moving the variety-matching to the cheapest place to do it. And the manager who demands "flexibility" without funding a single new response is asking for the one thing the law forbids — control without variety, absorption without a repertoire. Flexibility with no added moves is just a more anxious way of being out-varied.
The mapping
| In the model | In the world |
|---|---|
| disturbance states N | How many kinds of situation the world throws at you — the variety you must answer to. |
| regulator responses M | The size of your repertoire: distinct moves, forms, roles, settings you can actually deploy. |
| the acceptable set | What counts as staying in control — the narrow band of outcomes you're trying to keep the system inside. |
| M < N | Being out-varied: some states have no answer, and a fixed fraction of failures is guaranteed no matter the effort. |
| sensor noise ν | Variety lost in perceiving the problem — a response you can't aim is a response you don't have. |
| attenuation | Shrinking the world's variety upstream — standardise, filter, buffer — so that less regulation is required. |
Where it tears
The law says you can't keep every state acceptable when out-varied. It does not say you should try. Sometimes losing control of some disturbances is fine — even optimal — and matching every one is over-engineering. What the world needs is requisite variety, not maximal variety. A regulator that covers exactly the states worth covering, and lets the rest pass, can beat one that exhausts itself matching noise.
The model treats every disturbance as equally worth catching, because it counts states and nothing more. Real life is lopsided: a regulator that covers the rare-but-catastrophic and misses the trivial can vastly out-perform a bigger one that spreads its repertoire evenly. Before matching variety, weight it — a small table aimed at what matters beats a large table aimed at everything.
Every empire-builder claims the world just got more complex, so their department must grow to match. Requisite variety is the perfect alibi — which is exactly why the honest use of it is to audit the premise. Is the disturbance variety real, or manufactured? Half of good regulation is attenuating N down to what genuinely varies, so the answer is fewer responses, not more.