Hiring.

The framework

1. 01

   One question set, one order, every candidate.

   The comparison is the entire point. If candidate A got a warm chat about the industry and candidate B got a grilling on delivery, you haven't run two interviews — you've run two different tests and then ranked the scores against each other. Write the question set from the scorecard before the first CV arrives, ask the questions in the same order, and hold the follow-up probes to a written list as well. It feels rigid for about four minutes, which is roughly how long it takes to notice that you're finally hearing the same job discussed five times instead of five different conversations.

2. 02

   What they did, or what they'd do?

   Two families of question carry real evidence. Situational questions — Gary Latham and colleagues, Journal of Applied Psychology, 1980 — pose a dilemma from the job and ask what the candidate would do. Past-behaviour questions, formalised by Tom Janz two years later, ask for an occasion when they actually did it, then dig. A hypothetical rewards fluency rather than experience: asked what they'd do, a candidate answers from the best version of the job they've read about, and the more senior and rehearsed they are, the smoother it gets. A real incident has dates, a name, a decision that went badly — and all of that can be checked.

3. 03

   Score each answer against a written anchor — not the person against your impression.

   Patricia Cain Smith and L. M. Kendall gave the mechanism in 1963, in the Journal of Applied Psychology: write out what good and bad actually sound like, in observable behaviour, before anyone is interviewed, so the interviewer isn't asked to judge but to match. Latham's situational interview is where the familiar shorthand comes from — a benchmark answer written for a 1, a 3 and a 5 per question. Anchors do their work twice: they force you to agree what good looks like while nobody's face is attached to it, and they leave a record a year later when you want to know whether your standards were any use.

4. 04

   The contamination starts before the debrief.

   Everyone writes and scores alone, and nobody says a word until the scores are in — that rule belongs to the bar raiser's independent veto and to anchoring, and both of those mornings make the case better than this one can. Campion's list names the earlier leak, and almost nobody guards it: interviewers comparing notes between sessions. One 'you'll like her' in the corridor and the second interview stops being a second reading; it becomes a search for what the first person already found. Keep the panel identical across candidates, keep them apart until every session is done, and the four opinions you collect stay four.