What the noise at the time predicts

The result

Of the nine removals somebody argued about at the time, none later surfaced as a problem.

Of the seventeen nobody remarked on, four did.

The median wait before anything at all happened also sorts cleanly: four and a half weeks for the argued, six and a half for the questioned, eighteen for the unremarked.

The first objection to it

Nine and seventeen are small numbers, and one different outcome would soften the result considerably. If a single argued removal had surfaced, the headline would read "one of nine against four of seventeen", which is a much less striking sentence.

That objection is correct and it should be held throughout. Nothing here establishes a rate, and anybody who reports this as a percentage has over-read it.

What makes it worth stating anyway is the second column — the medians, which do not depend on the five rare events at all. Those are computed from twenty-four observations spread across every group, and they run in the same direction, monotonically, by a factor of four. Two independent features of the same record point the same way.

And there is a mechanism, which is the part that should carry most of the weight.

Why a contested removal reports quickly

An objection is not a prediction. It is an appointment.

Somebody who argued against losing a thing has, from that moment, a stake in its absence being noticed. They watch for the consequence — not maliciously, just as a matter of attention, because they said something would happen and they would like to know whether it did.

When the consequence arrives, three things are already in place. They notice it, because they are looking. They have standing to raise it, because they raised it before. And they are believed, because they said it would happen.

So the removal converts into information inside a fortnight. Either somebody asks for it back or nothing occurs, and either way the organisation has learned something it can act on.

Why a quiet one does not

An unremarked removal has none of that. There is no person who predicted anything, no one watching, and no one for whom the absence is a point already made.

It stops. Nothing happens. And the fact that nothing happens is read as evidence that the thing was not needed.

That inference is the error, and it is almost irresistible because it feels like evidence arriving. It is not. "Nothing happened" and "nobody is reporting" produce identical observations from inside, and on this record they were the same observation in four cases where the thing was very much needed.

Silence after a removal is not a result. It is the absence of a result, and the two look the same on a dashboard.

What makes something unremarked in the first place

Worth asking, because it is not randomness. Three things show up repeatedly in the seventeen:

It belonged to nobody. The standing slot for ownerless work had, by definition, no owner to defend it. Four other entries are of this type.

It lived in a slot rather than in a decision. Seven of the thirty-eight stopped because the hours they occupied stopped existing. Nobody could object to a proposal that was never made.

It had no visible output. A log that nobody read back, a reconciliation whose figures existed elsewhere, a contract reading that produced nothing when the terms had not changed. There was nothing to point at in the meeting, so nothing was said.

All three are properties that make a thing hard to argue for. None of them is a property that makes a thing unnecessary.

What this does and does not license

It does not license keeping the quiet ones. Thirteen of the seventeen are perfectly fine and several are among the clearest wins on the list.

It does not license dismissing objections. The nine argued removals include three of the fourteen things nobody has missed at all, which means the objectors were simply wrong three times out of nine and the removals were right to proceed.

What it licenses is narrow and cheap: when attention has to be rationed — and it always does — the quiet entries are where it buys the most, because they are the ones with no other reporter. The loud ones already have somebody watching, and that person will do the work for free.

What the twelve questioned removals did

The middle group is the least discussed and it behaves as the mechanism predicts, which is mild support for the whole account.

Of the twelve removals somebody asked about without objecting: three were asked back, four crept back, four were never missed, and one surfaced as a problem. The median wait was six and a half weeks.

A question creates a weaker version of what an objection creates. Somebody has shown enough interest to speak, which means somebody is mildly attending, but nobody has staked a position and nobody has a prediction to be vindicated. The attention fades faster.

That the middle group sits in the middle on every measure — lag, problem rate, how much came back — is not proof of anything. It is what a mechanism driven by the strength of somebody's attention would produce, and an inverted or flat middle would have been a reason to doubt the whole reading.