The interruption is not the cost

For about a decade my work had a shape that most work does not. Once a month there was a stretch of it that could not be done in pieces.

Reconciling accounts through a run of acquisitions is the clearest version. You are holding two records that should agree and do not, and you are working through the differences one at a time, and the differences are not independent. The fourth one you explain often unexplains the second. There is no state to save. There is no point at which the work is half done in a way that survives you standing up.

What you are actually holding

It is not knowledge. Knowledge would be fine, knowledge sits in a file.

What you are holding is a stack of provisional explanations. Seven items do not tie. You have a candidate reason for four of them, none of which you believe yet, and you are carrying the reasons alongside the confidence you have in each one, plus the order you intended to test them in, plus two things you noticed in passing that you have decided are probably nothing. None of that is written anywhere. Writing it down costs more than holding it, which is the whole reason it exists in that form.

An interruption does not pause the stack. It empties it.

When you come back you rebuild, and you rebuild imperfectly, because you can recover the seven items from the page but not the confidence rankings or the two things you noticed in passing. So you re-derive some of it, and you lose one of the provisional explanations, and forty minutes later you find it again and it feels like a new idea.

Invisible on one side, unmeasurable on the other

This is a genuinely difficult cost to talk about, and I think that is why nobody does.

The person interrupting sees a two minute question, and they are right. That is what it cost them. They have no way to see the stack, because the stack has no representation outside your head. There is no signal you can wear.

And the person interrupted cannot measure it either. You cannot say the question cost thirty five minutes, because you do not know, and if you produced a number you would sound precious. So the cost is real, borne entirely by one party, and never enters any account of how the work went. It surfaces only as a person who is more tired than the hours explain.

The part I got wrong

I became protective about it, and I stayed protective for longer than the argument justifies.

In delivery work across a lot of concurrent accounts, most of my value is being reachable. A question that waits four hours because I was defending a block of concentration can become a day of work built on a wrong assumption, and that trade is not close. The reconciliation instinct, applied to a coordinating job, is just expensive vanity.

There is a worse version of the correction, though. A meaningful share of the interruptions I resented were caused by me. Somebody had to ask because the thing they needed only existed in my head, and it only existed in my head because I had decided, weeks earlier, that writing it down would cost more than holding it.

So the stack I was so pleased to be able to carry was, at least some of the time, just undocumented work with a flattering name.

What is left of the argument

The cost is still real. It is just not owed to me by the people asking.

The version I believe now is smaller and less satisfying than the one I used to argue. Some work genuinely cannot be paused, and that fact is invisible to everyone around it, and the only honest thing to do about it is know which of your work is that kind and stop pretending the rest of it is.