Hiring.

The framework

1. 01

   The weights, written down, before anybody meets anybody.

   Decide in advance what each piece of evidence is worth. The work sample counts double; the structured interview counts once; the reference search counts once; the CV counts nothing, because it already did its job at the screen. The double weight is a choice rather than a ranking — the revised list puts the structured interview on top, and the sample takes the extra point here for a different reason: it's the one instrument where you watched the work instead of hearing it described. Swap them if you'd rather, but write the reason beside the weight. The number you put on the file before the interview doesn't enter this at all; that was a calibration check on the hour, never a column in the total. What matters isn't the cleverness of the numbers — it's that they were fixed while you were still impartial, because the moment you meet somebody charming the weights start moving on their own. The bar those totals get measured against isn't today's job. Day 18's Bar Raiser gave you that: a minimum written before the CVs, defended by somebody outside the team. Today is the number you carry to it.

2. 02

   Equal weights will do — the cleverness is in the list, not the arithmetic.

   This is the part people refuse to believe. Robyn Dawes showed in American Psychologist in 1979 that giving every predictor the same weight beats expert intuition, and that grinding towards optimal weights adds surprisingly little on top of that. So don't spend a fortnight calibrating. Spend it on the list: which pieces of evidence you'll actually gather, and which ones you're carrying out of habit because they were on the last scorecard. Then set them all to one, promote the work sample to two if you genuinely believe it, and go. A crude rule applied consistently outperforms a beautiful rule you abandon in the debrief.

3. 03

   What do you do when the rule is obviously wrong?

   Meehl anticipated this and gave it a name: the broken leg. A formula predicting whether a man goes to the cinema tonight should be overridden the moment you learn he's just broken his leg, because you hold a fact the formula could never see. Genuine broken legs turn up in hiring too — the reference who tells you something the scoring sheet had no column for, the visa that won't come through, the quiet admission that they don't want the job. What those have in common is that they're specific, checkable and new. A strong feeling about somebody in the room is none of the three.

4. 04

   An override with a name on it, or no override at all.

   The problem isn't that overrides are wrong; it's that people find broken legs constantly, and almost always in favour of the candidate they already liked. So make the override expensive. It has to be one specific fact, written into the record, with a name beside it, and read back a year later against how the hire actually turned out. That last step is what stops the exception quietly becoming the process — and what keeps the decision from reverting to its natural state, where the call goes to whoever holds the most authority or speaks last.