Synopsis.
Name
The Simpler Lever. Someone holds an outcome steady, and holding it steady is a real accomplishment: the world throws disturbances at the thing that must stay put, and for each disturbance the regulator has a response that cancels it, so the outcome does not move. And here is the quiet arithmetic underneath: to hold the outcome against many kinds of disturbance, the regulator must carry many kinds of response, at least one distinct answer for each distinct thing the world can do, because only a difference in response can absorb a difference in disturbance, and a regulator with fewer distinct states than the world has ways of disturbing it must, somewhere, meet two different disturbances with the same response and fail to cancel one of them. So the repertoire is wide because the world is wide. And then someone looks at the wide repertoire and sees, reasonably, cost and clutter: so many responses to maintain, so many categories to keep straight, so many branches in one rule, a menu too long to audit and too intricate to govern. And they simplify, and the simplification is locally right: fewer responses are cheaper to run, coarser categories are easier to read, one rule for many cases is simpler to teach and to check, and the trimmed lever is more legible, and a more legible controller reads as a better-governed one. And each cut lands cleanly, because in the moment of the cut the disturbances that needed the trimmed responses have not yet arrived, so nothing visibly breaks, and the console looks tidier and the governance looks tighter and the outcome is still, for now, held. And the repertoire keeps narrowing, one sensible economy at a time, each defended by its own legibility, until the count of distinct responses has fallen below the count of distinct disturbances the world can still produce. And nothing announces the crossing. The outcome stays held right up to the moment a disturbance arrives for which the trimmed regulator has no matching response, and then that disturbance passes through uncancelled, and the outcome moves, and no component has failed: the regulator is running exactly as designed, every response it keeps still works, it simply has no response for this, because the state that would have answered it was cut as clutter when nothing was arriving to need it. Nobody was careless. The simplifications were each rational, cheaper and clearer and easier to govern, and the lever read as better-governed at every step, more legible the more it was cut, and the outcome held the whole way down, so the trimming looked safe because a controller cannot show the disturbance it will fail to counter until that disturbance comes. It is only that a regulator can hold an outcome no wider than its own variety, that only variety absorbs variety, so when the responses are cut below the disturbances the control is already gone, silently, and shows only when the world does the thing the lever no longer has an answer for.
Arc
The bend is this: only variety can absorb variety, so a regulator holds an outcome only while it carries at least as many distinct responses as the world has distinct disturbances, and each simplification, locally cheaper and more legible, lowers that count until it falls below the world's, and the disturbances with no matching response pass through uncontrolled, the outcome slipping though no component fails. It runs on a regulator that holds an outcome by matching responses to disturbances. To keep an outcome steady the regulator meets each disturbance with a response that cancels it, so regulation is the pairing of a repertoire of responses against the variety of things the world can do, and stability is nothing more than having an answer ready for each (Ashby, on regulation and stability, the regulator that keeps an essential variable within bounds by countering each disturbance). And the count is the whole of it: only variety can absorb variety. A regulator with fewer distinct states than the environment has disturbances must meet two different disturbances with one response and leave one uncancelled, so to hold the outcome the regulator must command at least as much variety as it faces, the repertoire wide because the world is wide (Ashby, on the law of requisite variety, only variety can destroy variety, the regulator's variety setting the floor on the outcome it can hold). And the wide repertoire reads as cost. So many responses to maintain, so many categories to keep straight, so many branches in one rule: the variety that holds the outcome is exactly the variety that is expensive to run and hard to govern, so simplifying it is locally rational, cheaper and coarser and shorter, each cut a real economy (Beer, on variety engineering, the cost of proliferating variety and the standing pressure to reduce it). And the trimmed lever reads as better-governed. A shorter menu is easier to audit, a coarser category easier to read, one rule for many cases simpler to teach and check, so the simpler lever looks more legible, and a more legible controller looks like a better-controlled one, the legibility standing in for the regulation it is quietly draining (Beer, on the seduction of the legible control, a system made readable by throwing its variety away). And each cut lands without a sign. In the moment of simplifying, the disturbances that needed the trimmed responses have not arrived, so nothing breaks and the outcome still holds, and the repertoire narrows one defensible economy at a time until its count of responses falls below the count of disturbances the world can still produce (Ashby, on the requisite-variety inequality, the outcome's variety bounded below by the shortfall of the regulator's variety against the disturbance's). And a trimmed regulator is a smaller model of its world. To counter a disturbance the regulator must in effect model it, hold a state that corresponds to it, so cutting the repertoire cuts the disturbances the regulator can still represent and answer, and the ones it no longer models are the ones it can no longer regulate (Conant and Ashby, on every good regulator of a system being a model of that system, regulation succeeding only so far as the regulator models what it must counter). And the crossing shows only when the world does the untrimmed thing. The outcome stays held until a disturbance arrives for which no response remains, and then it passes through uncancelled and the outcome moves, and no component has failed: the regulator runs as designed, every response it keeps still works, it simply has no state for this one, the control lost silently at the cut and revealed only now (Beer, on the loss of requisite variety, the regulator overwhelmed not by breakage but by a variety it can no longer match). None of it is an error. Each simplification was cheaper and clearer and easier to govern, the lever more legible at every step, and the outcome held the whole way, so the trimming looked safe because a controller cannot exhibit the disturbance it will fail to counter until that disturbance comes, and a shortfall of responses is invisible exactly while the responses are not yet needed. Match the variety, and the outcome can be held again: amplify the regulator's repertoire so it carries a response for each disturbance once more, or attenuate the incoming variety with a filter that reduces the disturbances the regulator must meet, so the pair together commands the requisite variety: but amplifying means rebuilding the cost and clutter the simplification removed, and attenuating means adding a whole apparatus upstream and accepting the disturbances it screens off, and both read, against a lever that looked better-governed for being simple, as needless complication, so the outcome is held only by paying, continuously, to carry as much variety as the world can throw (Beer, on variety engineering by amplification and attenuation, the regulator's variety raised or the disturbance's lowered until the two match, at standing cost).
Stakes
Whether an outcome can be held steady, when holding it needs as many distinct responses as the world has disturbances and every simplification cuts the count. Every plot but the last moves the same way: toward the simpler lever, cheaper to run and easier to govern and more legible for being cut, which each economy is right to make, and away from the matched variety, which costs back the cost and clutter the simplification removed and reads as needless complication. The production cannot make a small repertoire hold a wide world, because only variety absorbs variety and a regulator holds an outcome no wider than its own count of responses, so a repertoire cut below the disturbance variety leaves some disturbances with no matching state and the outcome slips, not through any fault: no component fails and every response kept still works. It can only show that each cut was a real economy and the lever read as better-governed the more it was trimmed, that the control was lost silently at the cut and not loudly at the slip, and that the one move which holds the outcome, matching the variety by amplifying the repertoire or attenuating the disturbance until the pair meets the law, must be made by paying, continuously, to carry as much variety as the world can throw and to accept as complication what read, simplified, as good governance, so the lever is legible and the governance looks tight and the outcome it was meant to hold is the one thing it can no longer hold.
Estate
khai-misfits: the house that holds this production and answers for the whole run.