Not the same question

An argument that does not close

The return-to-office argument has been running for five years. In that time it has accumulated a large research literature, several natural experiments and an enormous quantity of opinion, and the positions in it have not converged. The people who thought offices mattered in 2021 think so now. The people who thought the opposite think the opposite. Individuals change their minds, usually when their own circumstances change; the argument as a whole sits where it started.

That is unusual. Most disagreements about how organisations should work either narrow under evidence or get settled by somebody and stop. This one does neither, and the explanations usually offered — that one side is self-interested, that the evidence is mixed, that it is really about control — all predict something that does not happen. Self-interested parties bargain. Mixed evidence produces narrowing. A dispute about control produces a negotiation.

What actually happens is that both sides keep producing studies.

This site offers a different explanation, built on one organisation's record, and it is mechanical rather than moral: the argument contains three different kinds of claim, and they are being traded against each other as though they were one kind.

Three kinds

Ask of any claim in this argument a single question — what would have to be true for this to be settled? — and the answers fall into three groups.

Empirical. Something measurable would have to come out one way rather than another. "New starters take longer to become productive remotely" is this kind. It might be true or false; what matters is that the world could tell you.

Value. Nothing measurable would settle it, because it is a statement about what matters. "Work should be something you go to" is this kind. So is "people should be free to choose where they live". Neither is weak, unserious or a smokescreen; they are a different sort of thing from the first group, and no amount of evidence improves or damages them.

Interest. It is settled by naming who bears a cost. "The lease runs to 2031" is this kind, and so is "the travel falls entirely on the employee". These are the shortest claims in the argument, the cheapest to establish, and — on the record counted here — the least often said out loud.

Each kind has its own terms of settlement and they do not transfer. Evidence settles the first and is inert against the second. A principle can answer the second and cannot establish the first. A statement of cost does something neither can do, and costs almost nothing to make.

What was counted

One organisation. One consultation about a proposed three-day requirement, over eleven weeks: a proposal document, two all-hands sessions, a written consultation with ninety-six responses and three team meetings.

Every claim made in support of or against the proposal was written down in the words it was said in. Twenty-four of them, each counted once however many times it was made. Each was then sorted by kind, and recorded against what it was answered with.

Nine were empirical. Ten were about values. Five were about who pays.

What it came to

Twelve of the twenty-four were answered with a kind of answer that cannot settle them.

Eight times, somebody said what they thought mattered and was handed a study. Four times, somebody made a checkable claim and was handed a conviction. In every one of the twelve the reply was true, on-topic and delivered as a rebuttal, and in every one of the twelve it bore no relation to the claim it answered.

Five were never answered at all — four of those because they were never said. All four were costs, all four were known to everybody involved, and between them they were the decisive input into the decision.

Seven were answered in their own kind, and those seven produced everything useful that happened in eleven weeks: two trades, one measurement, and one disagreement recorded as a disagreement rather than as a mistake.

Seven exchanges out of twenty-four were capable of going anywhere at all.

The empirical part is smaller than it looks

Both sides experience this argument as a dispute about facts. On the record here, a little over a third of it is.

Of the nine empirical claims, two could have been settled with data the organisation already held. One of them was — it took an afternoon, and it is the only claim out of twenty-four that was both empirical and answered with evidence that measured it. The other was never checked, by anybody, in eleven weeks.

Three were answered with studies that measure something else: a paper comparing companies produced at a claim about teams, revenue per head across a period in which the product changed, sick-pay behaviour from a country with different rules. All three citations were made in good faith by people who had read the papers. Reading a paper and checking that it addresses your claim are different acts, and only the first feels like diligence.

And for four of the nine, no evidence exists anywhere, for or against. Nobody records how much time colleagues spend together. Nobody traces ideas back to where they started. Nobody has written down what the office actually did, which means the claim that tools have replaced it cannot be wrong — which is not the same as being right.

So even if every available study were gathered and read perfectly, roughly four fifths of this argument would still be standing. That is why more evidence has not helped and will not.

The claims nobody made

The strangest thing in the record is not that the costs were unevenly distributed. It is that four of the five were known to everybody and mentioned by nobody.

The lease, which runs to 2031 with a rent review next year, was stated once — in a finance update, to a different group of people. The travel, which ranges across this team from eighteen minutes to an hour and twenty-five each way, was never stated. Neither was the childcare built around a pattern now being changed, nor the two house moves made on the basis of the previous arrangement, nor the fact that the people writing the rule had the shortest journeys on the team.

Four silences, four different reasons, all of them predictions about how the statement would land. The travel would sound like a platitude. The childcare would mark somebody as less available. The house move would sound like asking for an exception. And the lease would make the collaboration argument sound like a cover story.

On the evidence of this record the collaboration argument was not a cover story — after the decision was taken, with nothing left to secure, it went on being made in the same terms. But a sincere argument standing next to a financial constraint looks like a pretext, and knowing that, people leave the constraint out. What results is a decision that everybody experiences as having been made on the merits and that nobody can reconstruct, because the merits as recorded do not add up to it.

Nobody here was arguing in bad faith

The accusation runs in both directions and it is the most popular explanation for the deadlock. It is also, on this record, wrong.

If people were arguing in bad faith you would expect the cover story to be dropped once the decision was secure. It was not. You would expect claims to be abandoned quietly when refuted and replaced with new ones aimed at the same conclusion. That did not happen either. What happened was simpler and harder to fix: people reached for the most persuasive register available, which in any meeting is the empirical one, and they did not know which kind their own claims were.

Asked afterwards, several participants were surprised to be told that their central claim was not empirical. One disputed it for ten minutes before agreeing.

That matters practically, because assuming bad faith is expensive. Once you believe the other side is insincere, their value claims become tactics, so you stop answering them — which is the one thing that could have moved anything.

Both sides do all of it

This is not a method for showing that one side argues badly. On this record each side made claims of all three kinds, each reached for the empirical register beyond what it could support, and each answered the other in the wrong kind at about the same rate.

Of the eight studies produced at value claims, five came from the side arguing for the office and three from the side arguing against. Both sides left their own costs unstated. Both sides had one claim so generally phrased that no observation could contradict it, and both reached for it at the moment they felt they were winning.

There is one asymmetry that is real and it is not about conduct: the costs are not symmetrically distributed or symmetrically visible. Four of the five fell mostly or entirely on individuals, and the one that fell on the organisation is the only one anybody mentioned. That follows from where costs are recorded rather than from anybody's character, and noticing it is not taking a side.

What this site does not do

It does not say how many days anybody should work in an office. That question contains claims of all three kinds and a website does not settle it; the site's whole argument is that such a question is not the sort of thing that gets settled by producing more material.

It does not say who is right. Sorting a claim tells you what would bear on it and nothing about whether to accept it. An empirical claim correctly identified may be false; a value claim correctly identified is not thereby defeated, and the sentence to avoid is "that is just a value claim" — the damage is in the word just.

It does not neutralise power. A disagreement accurately located is still resolved by whoever decides. What it offers the person being overruled is the difference between being overruled on their actual position and being overruled on a position nobody understood, which is smaller than winning and larger than nothing.

And it rests on one record: twenty-four claims, one organisation, eleven weeks, sorted by one person. Two of the twenty-four are arguable in their kind and are flagged where they appear. The method may travel. The numbers do not.

What is worth doing

The whole technique is a four-second pause before replying, and one question asked silently: what would have to be true for this to be settled?

The answer determines the reply. To an empirical claim: is there anything we could look at? To a value claim: that is a real difference between us, and here is mine. To a claim about cost: what is the figure, and what is mine?

None of the three requires the other person to know anything about any of this, and the reliable effect is that people answer in the kind they are handed. A value stated plainly invites a value back. A measurement named invites a question about the measurement. A figure given invites another figure.

In eleven weeks, across six sessions and twenty-four claims, nobody in this consultation took that pause once. The sorting is not difficult — almost everybody gets it right on the first attempt when they stop to do it, including on claims they strongly disagree with. It is simply never done.

How this is arranged

Six parts. The first sets out the three kinds and the question that tells them apart. The second is about why the argument does not close, and what a closed one would look like. The third takes the nine empirical claims and asks what would settle each. The fourth is about the twelve crossed answers — what each crossing is, why the wrong answer feels like an answer, and how to catch yourself doing it. The fifth is the five costs and the four silences. The sixth is what to do with a claim once you have sorted it, which is three sentences and a habit.

Every figure anywhere on this site comes from the same twenty-four claims. No person, employer or workplace is named, and none will be.