How long until anyone finds out

The range

On thirty-eight removals, twenty-four produced an event inside seventy-eight weeks. The lags of those twenty-four ran from one week to sixty-four.

That is a span of sixty-three weeks on the same list, in the same organisation, among removals that looked, on the day they were made, like broadly the same kind of decision.

The fourteen that produced no event are not a fifteenth category of short lag. They are removals whose lag is either longer than seventy-eight weeks or infinite, and nothing in the record distinguishes those two cases. This matters: not missed is a statement about the observation window, not about the thing.

Why the lag is the useful quantity

The usual question about a removal is whether it was a mistake. That question has the disadvantage of not being answerable at the time, and the further disadvantage that by the time it is answerable it is no longer a question.

The lag is answerable in advance, roughly, and it tells you what to do. Asked of any removal: how long before we would know?

• A lag of days or weeks means the removal is self-correcting. Make it, and the organisation will tell you.
• A lag of months means the removal needs a date in a calendar.
• A lag you cannot estimate is the dangerous case, and it is dangerous in a specific way: an unestimatable lag is usually long, because short lags are easy to see coming.

Three things the lag is not

It is not severity. The fastest event on this record was a request for printed rotas, one week after they stopped, and it cost nothing. One of the slowest was a log whose absence caused the same piece of work to be done four times. There is no relationship between how quickly a removal reports and how much it matters, and assuming otherwise is the commonest error in reading a list like this.

It is not importance at the time. The removals that produced the fastest events were, in the main, the ones somebody had objected to. Objections produce speed, not significance.

It is not a property of the organisation. The same organisation produced one-week lags and sixty-four-week lags in the same quarter. The lag belongs to the removal — specifically, to how the absence of that particular thing would make itself known.

What sets the lag

Reading the twenty-four events against what each removal had been doing, the lag appears to be set by one thing: who is positioned to notice, and how often they are in a position to notice.

A printed rota is noticed by whoever reaches for it, which is the next morning. A monthly stock count is noticed at a month end. A contract read before renewal is noticed at a renewal, which on this record was twice a year. A log of refusals is noticed when the same refusal comes round again, which could be any interval at all and happened to be fourteen months.

The lag is not how long the consequence takes to occur. It is how long until somebody is standing in the place where the consequence is visible.

This is why the lag can be estimated at the time of the removal, without knowing anything about whether the removal was wise. You are not predicting the consequence. You are asking when the next occurrence of a known event is.

The practical form

Three columns, filled in when the removal is agreed:

1. What stopped.
2. Who would be the first to be standing somewhere the absence shows.
3. When that next happens.

The third column is a date. If it cannot be filled in, that is the answer: the lag is long, and the removal needs a check that does not wait for the lag to run.

On this record the third column was not filled in for any of the thirty-eight. Had it been, four of the five costly removals would have had a date in it that fell inside the first six months — a quarter end, a renewal, an induction, a year end — and the fifth would have had nothing, which would itself have been the signal.

Estimating a lag without knowing anything

The claim that lags can be estimated at the time deserves a test, and the record allows a weak one.

Reading the thirty-eight entries with the outcomes covered, and asking only the question above — who is first positioned to notice, and when — produces a date for twenty-nine of them. For most the honest answer is "next week", which turned out right: twenty-two of the twenty-four events arrived from somebody whose own work was affected.

Of the nine where no date could be given, three were the three slowest failures.

That is nine entries flagged out of thirty-eight, containing three of the five problems, from a question asked without any knowledge of the outcome. It is not a prediction of which removals were mistakes — six of the nine were fine — but it is a usable filter, and it costs a sentence per entry.