The quiet ones
What this does not license
The finding on this site is easy to over-read in four specific directions, three of which would make a transition worse.
The quarterly reconciliation of the project ledger
week 3 · surfaced after 31 weeks
Shadowing new starters through their first week
week 6 · surfaced after 47 weeks
Reading supplier contracts before renewal
week 2 · surfaced after 22 weeks
Logging why a job was turned down
week 9 · surfaced after 64 weeks
Surfaced 4
Four removals were written down with the week each of them stopped. Every one of them produced an event inside the 78 weeks watched. Of these, four surfaced as a problem. The four removals nobody said a word about, all of which later surfaced as problems.
A result that sounds counter-intuitive invites over-reading, and this one has four obvious over-readings. Three of them would produce a worse transition than paying no attention at all.
It does not license keeping the quiet removals
Seventeen removals passed without comment. Thirteen of them produced no problem at all, and several — the duplicate manager copies, the unread logs, the deck nobody presented — are the clearest savings on the record.
If the rule were "a removal nobody argues for is suspect, so keep it", the organisation would have retained thirteen useless procedures to avoid four problems, and the shorter week would not have been available at all, since the hours had to come from somewhere.
The finding is about where to look, not about what to do.
It does not license overriding objections
The symmetrical error. If argued removals reliably turn out fine, why not treat an objection as a positive signal and proceed with confidence?
Because the four fastest useful corrections on this record came from objectors, and because in two of those four what came back was substantially better than either the original procedure or its removal. Objections produced the best outcomes on the list.
More to the point, the mechanism described here depends on objections being taken seriously. An objector watches and reports because reporting is worth something. In a place where objecting is pointless, people stop doing it, the nine fast reports become nine silences, and the organisation has converted its best-behaved category into its worst.
The finding is parasitic on objections being respected. Using it as a reason to disregard them destroys it.
It does not license predicting which removals will matter
Shown this record without the outcome column, people pick the contested ones. That is the error the whole site is about and reading the site does not inoculate anybody against it.
Nothing here supports a claim that quiet removals are likely to be problems. Four of seventeen became problems — which means thirteen did not, and the base rate of trouble among the quiet ones is still low.
The claim is about reporting, not about risk. A quiet removal is not more dangerous. It is less observable, and the remedy is observation rather than caution.
It does not license treating this as a rate
Nine and seventeen are small. Five problems is smaller. A single different outcome in the argued group would change the headline from "none of nine" to "one of nine" and the sentence would lose most of its force.
Anybody quoting this as a percentage, or applying it to a transition of four hundred removals, has taken it past what it can carry. The medians are on firmer ground — twenty-four observations rather than five — and even those are one organisation's quarter.
What travels is the mechanism and the two questions. The numbers stay here.
What it does support
Four things, all narrow:
- Record whether anything was said, at the time, in three values. It costs seconds and it is the only field on this record that predicted anything.
- Ask who in the room has done the thing, before agreeing to remove it. Catches the specific case where silence is absence.
- Ask what people used to do on the day that went, a month afterwards. Catches the removals nobody proposed, which are invisible to every other method.
- Read the quiet entries back twice a year, with a date beside each. Twenty minutes, and it is the only activity aimed at the population that needs it.
None of these says anything about whether a shorter week is a good idea, and none of them argues for or against any particular removal. They are all about the same thing: making sure the organisation finds out what it did, at a time when finding out is still cheap.
What would change the conclusion
Worth stating, since a finding that cannot be contradicted by anything is not a finding.
A second record where the argued removals produced the problems. That would kill the mechanism outright rather than just the numbers, and it is the obvious test.
A record where the medians do not sort. The four-and-a-half, six-and-a-half, eighteen progression is the firmer half of the evidence here, because it rests on twenty-four observations rather than five. If quiet removals reported as fast as loud ones elsewhere, the account would be wrong even if the problem counts happened to repeat.
An organisation where objecting is costly. The mechanism predicts that the fast reporting disappears there, because objectors stop objecting. If a place with suppressed dissent still showed fast reporting on contested removals, something other than attention is doing the work.
None of those can be settled from this record, which covers one transition. They are the shapes a disagreement with it could take, and anybody with a second list is better placed to look than anybody with an argument.