A Rule Is Relief From Ambiguity

It is 22:40 on a Thursday and you are writing a rule.

The week produced the reason. A quote went out with a discount nobody had approved, or a contractor shipped work that did not match what was promised, or a customer heard two different answers from two people in the same afternoon. Small things. Each individually survivable. Together they left a residue that has been sitting behind your sternum since Tuesday, and it has a specific texture: not fear exactly, more the low hum of a question that will not close.

So you open the shared document and you write. From now on, all discounts above ten per cent require sign-off. Three sentences. You read them back, adjust one word, and save.

And something settles.

Notice what has just happened. Nobody has read that rule. Nobody has followed it. No discount has been approved or refused under it. Not one thing about your company has changed in the last four minutes. The document is not even circulated yet.

Something about you has changed, though. The hum stopped.

A rule is the only management act that pays out before it is tested.

The payoff arrives before the verdict

Every other decision you make carries its verdict into the future, and makes you wait for it.

You hire someone and you find out whether you were right across the following four quarters. You reprice and you learn what it cost you when renewals come round. You commit to a market and the answer arrives, if it arrives cleanly at all, somewhere on the far side of two years. This is most of what the work is. You act into partial information and you carry the not-knowing until reality gets back to you.

The rule does not behave like that.

The rule pays at the moment of writing. Before circulation, before adoption, before the first case that tests whether the wording survives contact with a situation you did not imagine. The transaction completes inside your own head, at 22:40, and it completes in full.

What it pays out is the end of an open question. Before: I do not know how we should handle this. After: we handle it this way. That transition is the entire product. Be precise about what was purchased, because it was not a better outcome.

And be precise about why the rule is alone in this. Plenty of things you write down feel like progress the moment you write them - a target, a quarterly objective, a number announced to the team. But those open a loop rather than closing one. Having set the target, you now carry it. The rule is the only one that runs the other way: you write it, and you are carrying less than you were four minutes ago. The chance of a better outcome, maybe, eventually, entangled with a hundred other causes. What was purchased and delivered on the spot was the closing.

This is a described phenomenon rather than a metaphor. Psychologists call the underlying motivation the need for cognitive closure: a desire for definite knowledge on a question, and an aversion to confusion and ambiguity. Kruglanski and Webster’s 1996 formulation in Psychological Review is the standard reference, and the useful part is the split into two tendencies. The first is urgency, the inclination to reach closure as soon as possible. The second is permanence, the inclination to hold onto it for as long as possible.

Hold onto both of those. The first one explains tonight. The second one explains your handbook.

One further note, because it changes who this piece is about. Need for closure is not only a stable personality trait. It is also a situationally evoked state, which means it rises under conditions rather than sorting people into types. Fatigue raises it. Time pressure raises it. So does the residue of a week in which several things went slightly wrong at once.

Thursday, 22:40, after that week, is not a personality. It is a condition.

Who takes the relief, and who pays for it

Trace where each half of the transaction lands. They land in different places, and that separation is the mechanism.

You take the relief. All of it. It arrives inside the person doing the writing, at the moment of writing, with no lag and no discount for uncertainty.

The cost goes somewhere else. It goes to whoever now has to route a decision that used to take four seconds. It goes to the person who can see the discount is obviously right for this account and now needs your attention to say so. It goes to the salesperson who stops making close judgement calls because there is a procedure, and who is therefore fractionally worse at making them a year from now than they were the day you wrote it. It goes into the gap between when a customer asks and when the company can answer.

None of that comes back to you as feedback about the rule.

It does come back, but transformed. It returns eight months later as a general impression that the company has become sluggish, that people are waiting for you more than they used to, that things which ought to be simple are taking a week: an unattributed sensation about the organisation, entirely detached from the twenty-three individual evenings that produced it.

And how do you solve a company that has become sluggish and unclear? Late one Thursday, with a rule.

You are not looking at somebody else’s disease, and you are further along the curve than the story you tell yourself about being lean would suggest. The version of this that gets written about is the enterprise version, and it is easy to read the enterprise version and conclude it describes a scale you do not run. It describes a destination you are already travelling towards, one Thursday at a time.

The relief is yours. The cost is theirs. Neither of you can see the connection.

The objection, at full strength

There is an obvious response to everything above, and it is correct.

Some rules are not optional. If you handle payment data there are things you must document. If you have institutional customers, their procurement process will require written procedure before it will require anything else. If you sell into regulated sectors, an auditor will one day ask to see the policy, and we use judgement is not an answer that survives that room.

And in August 2026 this got sharper for almost everyone. Article 14 of the EU AI Act became applicable on 2 August 2026, under the Act’s own Article 113. It requires that high-risk systems be designed so that they can be effectively overseen by people while in use, and it requires that whoever is doing the overseeing can understand what the system can and cannot do, notice when it malfunctions, resist over-reliance on its output, override it, and stop it. In practice that means written procedure. Documented. Reviewable.

So an operator reading this can say, with total honesty: I am not soothing myself. I am complying.

I take this objection seriously for a reason that is not rhetorical. I spent twenty years inside Nokia Business Infrastructure, in SOX compliance, risk management and process development. I did not observe the guardrail industry from outside it. I built guardrails, professionally, at a scale where they were the correct answer, in an environment where a missing control was a genuine liability and everybody involved knew it. I am not going to write a piece that treats process as a failure of nerve. Process is load-bearing, and I have seen what its absence costs.

Which is exactly why the objection does not weaken the argument.

It supplies it.

Two properties, neither of them in the text

A compliance rule and a rule written to settle your own discomfort are indistinguishable by reading them. They use the same words. They both look reasonable. This is why the natural test fails, and the natural test is is this rule necessary? You will answer yes. You will answer yes every time, because you are answering with the same faculty that wrote the rule, drawing on the same memory of the same week, and that faculty already returned a verdict at 22:40. Surveying your own handbook and finding it all defensible is not evidence. It is the same act repeated.

The two kinds of rule differ, and they differ structurally, on two properties of the record around them.

A compliance obligation has both by construction. That is what makes it compliance rather than preference: somebody else wrote it, somebody else checks it, and it comes attached to a review cycle. Article 14 is not a rule you invented on a Thursday: it has an author with a legal basis, and somebody whose job is to look.

A rule written for relief has neither property. It was authored by a feeling, and the feeling has long since been forgotten, which is why nobody in your company can tell you where the rule came from. And it is audited, for as long as it exists, by the person who wrote it. Who is not looking, because looking at it would reopen the thing it closed.

Run the two questions across your own handbook and you will find they sort it fast. Not into good rules and bad rules. Into rules with a provenance and rules without one.

The measurement: what you removed

The two questions are the test. This is the measurement.

Return to Kruglanski and Webster’s second tendency. Permanence: the inclination to maintain closure for as long as possible. Urgency explained the writing. Permanence explains what happens afterwards, which is nothing.

A genuine control system has a removal path. Rules that expire on a date. Rules that shrink as evidence accumulates. Rules that get deleted when the condition that produced them stops existing. Deletion is routine rather than an event, because a control system is a response to conditions, and conditions change.

Relief has no removal path, and the reason is structural rather than lazy. Deleting a rule reopens the question that rule was written to close. Deletion costs exactly what writing paid, only in reverse, and it charges that cost today for a benefit that is diffuse and unprovable. So it does not happen. The corpus accumulates, and every individual decision not to delete is locally reasonable.

There is a competing explanation here and it deserves saying out loud, because it is the better one at national scale. Rules might survive not because deleting them forfeits a reward, but because deletion is expensive in a different way: somebody has to justify the removal to somebody else, and nobody is thanked for that. Coordination cost, not foregone relief. Against a parliament, that account plainly beats mine.

It does not survive your company. Coordination cost predicts that deletion is easy wherever there is nobody to justify it to. You can strike a rule from that shared document this afternoon, alone, with no meeting and no memo, and nothing in your calendar would register it. The cost is nearly zero and the deletion still does not happen. Whatever is stopping you is not the meeting you would have to hold.

Do not count your rules, then. The count tells you nothing: a genuinely complicated business has many, and so does a fearful one.

Count the difference.

Over the last twenty-four months: how many did you add, and how many did you remove?

Rules, policies, approval thresholds, mandatory formats, standing checks, required sign-offs, the things in the handbook and the things that only live in what everybody knows you expect. Two numbers. Added, and removed.

Nobody legislates a duty that already feels good

The strongest evidence for this is not a statistic. It is a remedy that had to be invented.

On 1 January 2011 the United Kingdom became the first OECD country to adopt regulatory offsetting as official government policy: a one-in-one-out requirement, under which no new regulation could be introduced unless an existing one of equivalent or greater burden was removed. It was tightened to one-in-two-out in 2013, then to one-in-three-out, and given a statutory footing in 2015. Eighteen OECD countries had offsetting policies in place by 2024, and the European Commission runs a one-in-one-out approach set out in its 2021 Better Regulation Communication, which it describes as a cost brake rather than a mechanical swap.

Read what that means rather than what it says. Institutions that could see their own rule corpora growing, that had every analytical resource available to them, and that genuinely wanted the growth to stop, could not stop it by intending to. They had to pass a rule that forced them to delete rules. They had to give the deletion an external author and an external auditor, because on its own it was never going to happen.

Nobody legislates a duty to do a thing that is already rewarding.

Notice also what the escalation says. One-in-one-out was not enough, so it became one-in-two-out, then one-in-three-out, and then it needed statute behind it. Each step is an admission that the previous step had not produced deletion.

And one honest complication, because leaving it out would be its own kind of relief: in 2023 the United Kingdom repealed the statutory target altogether. The rule about deleting rules was itself deleted. That is not a joke at anyone’s expense. It is the same mechanism one level up, and it should tell you how much structure this actually takes.

What would mean you do not have this problem

I have to be exact about the status of this claim, because I would rather it be falsifiable than impressive.

Nobody has measured the addition-to-removal ratio inside operator-led companies. Not in anything located while researching this piece. The offsetting evidence is national-scale, and a government is not a fifteen-person company. What I have is a mechanism with a sixty-year literature behind its components, one domain in which deletion demonstrably had to be constructed rather than left to intention, and a prediction. I do not have a study of you.

So the count is a real test, and it can come out the other way.

Three results would mean this mechanism is not operating in your company.

If you removed rules at anything like the rate you added them, the permanence half of the argument does not describe you. If your heaviest rules sit where outcomes vary most and the stakes are highest, rather than where the most vivid recent incident happened, you are weighting by expected cost, which is governance doing its job. And if you can name the author and the expiry for most of what is in the handbook, then those rules were written by something outside your own discomfort and the mechanism has no purchase.

Any of those, and you are looking at a control system. Go and do something more useful with your Saturday.

You did not resolve it. You moved it.

The rule did not remove the uncertainty. The uncertainty is a property of the world. A document you wrote cannot touch it. Whether a fourteen per cent discount on this particular account is the right call remains exactly as unclear as it was before you had a policy about it.

What the rule removed was your contact with it.

The open question is still open. It has simply moved from you, who could have decided it in four seconds with full context, to whoever now has to apply a prescription written by someone who is not in the room, who will never see the case it does not fit, and who is not available at the moment the customer is waiting. And it moved to a person with less authority to reopen it than you had.

You did not resolve it. You relocated it, downward.

The reading underneath all of this

There is a layer below the mechanism, and it is what makes this so hard to catch from the inside.

The feeling of handled is real. It is fast, and it is the only immediate feedback you get on a decision of this kind.

But look at what it is actually responding to. It fires on the act of writing. It does not know whether the rule will work, whether anyone can follow it, whether the case that prompted it will ever recur, or whether it will still make sense in a year. It fired at 22:40 and it would have fired just as cleanly if you had written something incoherent.

An instrument that reads resolved the moment you touch it is not measuring the system. It is measuring your hand.

And an executive who steers by that reading, in good faith, with a defensible justification for each individual step, will over several years construct an organisation that nobody can move. Including them.

Where this sits

If you have read The Ungovernable Company, the two arguments may look like they point in opposite directions. That piece says build the governance structure. This one says you keep writing rules to settle yourself. Both are true, and they name opposite failures of the same system.

The Ungovernable Company is about a missing trail: the company cannot reconstruct how it decided, so nobody can govern, value or audit it. This piece is about proliferating prescriptions: the company has replaced deciding with lookup. A single company can carry both at once, and many do - a thick handbook and no recoverable reasoning behind any of it.

The reconciliation is in what each one asks you to build. A record of reasoning is not a rule. A rule is not a record. And notice that the four fixes that piece prescribes - a named owner, an aggregate decision trail, a preserved channel for dissent, a record kept diligence-ready - each arrive with an owner and an audit attached. They pass the two questions above. That is not a coincidence. It is the same standard, applied from the other side.

There is a closer sibling. The Substituted Question examines the same reassuring yes from the opposite direction: there the question you asked silently becomes an easier one, and you answer the substitute without noticing the swap. Here the question does not change at all. You ask the right one, answer it honestly, and the instrument you answer with was already paid at the moment you wrote the rule. One failure is in which question got answered. This one is in who answered it. The repairs differ accordingly: that piece asks you to fix an observable before you begin, this one asks you to audit a record you already have.

Two questions and a weekend

Take the handbook, and everything that functions as a rule whether or not it is written down. For each one:

Who authored this, other than my own discomfort? And who audits it, other than me, and when does it expire?

Then the count. Added, removed, twenty-four months.

Neither exercise requires a consultant and neither can be done wrong. They will produce a specific, unflattering, checkable answer, and the answer belongs to you before it belongs to anybody else.