File 03 — Judgment
The Unanimity Filter
Twelve reasonable people, one veto each. The average never moves. Both ends disappear.
Nobody can appoint you alone; almost anybody can end you alone. Put a number on that and the usual complaint — that the people at the top got blander — turns out to be the wrong claim. The average does not fall. The spread collapses. Set the number of vetoes, how readily any one of them objects, and how much is on an average candidate’s file, and watch a field lose both ends at once: fewer disasters, fewer people worth remembering, always in exactly the same proportion, because an objection does not come labelled with which way the person was going to turn out. Every figure is a closed form, not a simulation — the survivors of a Poisson filter are Poisson again, which is why the middle can be proved to stay put. Built on the German essay this argument comes from.
Nobody can appoint you alone. Almost anybody can end you alone. Set how many people hold that kind of veto, how readily any one of them objects to any one thing on your file, and how much there is on the file of an average candidate. What comes out is not a worse average. It is a narrower one — and the arithmetic cannot take a slice off the bad end without taking the same slice off the good end.
filled: the people who got through. dashed: the same field before anybody objected. same axis, same middle, same total — the columns only grow because the ends went away, and the middle one is allowed to run off the top rather than flatten everything else.
Twelve reasonable people
The local branch, the list, the caucus, the committee, HR, the board, compliance, the press office, an editor, a comment thread. None of them can appoint you. Each of them can end you.
gets through at all
5.5%
of everyone who started, once every veto has been asked
of the survivors, nothing on file
91%
not one item anybody could object to cleared
of the spread survives
18%
the middle is exactly where it was; only the width changed
the bill, per thousand appointments
runs it into the ground
112 → 2
every one of these is a story somebody can tell
gets remembered
112 → 2
nobody can name a single one of these, ever
The two numbers are the same number, and they always will be. The filter measures whether anyone objected, and an objection does not come labelled with which way the person was going to turn out.
So the process just bought insurance that works: 98% of the disasters are gone. It paid for them with 98% of the other end. The premium was never on an agenda, never in a report, and cannot be — a prevented catastrophe leaves a competitor you can point at, and a prevented great appointment leaves nothing at all.
now the candidate in front of you — everything here is true, and every objection to it is reasonable
3 things on file, 12 vetoes to clear: 1 in 31,500 chance of getting through.
That is the strict reading, where each veto forms its own opinion. If they all read the same file and object to the same things, the whole process collapses into one gate and the chance rises to 42%. The truth is somewhere in that band — and note which end of it requires the gatekeepers to be thinking independently.
Where this comes from. The veto mechanism, the six lines on the file, and the twelve-gate figure are from Charakter ist ein Risiko, und wir haben es wegversichert (in German). Its one-line version — twelve gates at a quarter each leaves 3% — is what this tool shows at a file of exactly one item. The population layer, the closed form, and the two-tailed bill are the arithmetic underneath it.
The honest catch. Three of these dials are assumptions, not measurements: nobody has counted the real vetoes in a real appointment, or the objection rate, or how much is on an average file. Change them and the numbers move a lot. Two things do not move: the middle stays where it is at every setting, and the two ends always shrink by the same factor. The model also assumes an item on your file is as likely to make you great as to make you dangerous — which is plainly not true of any particular item, and is exactly what a selection process that measured character rather than objections would be able to tell apart.