Picture the debrief. Five interviewers, one conference room, an hour blocked at the end of the day. Someone speaks first. Someone speaks loudest. Someone has the title that makes the others wait. By the time the quietest interviewer offers the observation that should have changed the decision, the decision is already leaning, and they revise their read to match the room.
This is what unstructured discussion does to good people. It rewards confidence and seniority, and it treats those as proxies for signal. They are not. The most certain person in the debrief is often certain about the wrong thing, and the room has no instrument for noticing.
The fix is to change the sequence. Score first, talk second, and build the score on a framework you agreed to before you ever met the candidate.
The rubric comes before the search
We build a weighted rubric for every role before sourcing begins. Not after the first few interviews, when the criteria have quietly been rewritten to fit whoever impressed you. Before. The weighting is the argument: it forces the hiring team to decide, in advance and on the record, what actually matters for this role and how much.
Calibrating up front does something subtle. It moves the hardest conversation (what are we really hiring for) to the moment when no real candidate is on the line. You can disagree freely about an abstraction. It is much harder to disagree honestly once a name and a face are attached, because now the debate is about a person and everyone knows it.
Two dimensions, scored separately
We score on two dimensions, and we keep them apart on purpose. The first is judgment and behavior: how the person reasons, how they handle ambiguity, how they work with people who disagree with them. The second is skill and depth: whether they can actually do the work to the standard the role demands.
Held separately, the two dimensions stop contaminating each other. A candidate with warm, fluent judgment can otherwise mask a thin technical core, and a deep specialist can lose a room that never probed how they collaborate. When you score the dimensions together, the louder one wins. When you score them apart, you see the shape of the actual trade-off you are being asked to make.
Score first, talk second. The discussion should explain the scores, not produce them.
Blind submission, then discussion
Every interviewer submits their scores before the group talks. Blind. No previews in the hallway, no read on which way the senior person is leaning. By the time the conversation starts, the independent signal is already captured and cannot be quietly revised to match the room.
This is the whole move. The value of five interviewers is that they are five independent measurements. The moment they hear each other before recording, you no longer have five measurements. You have one measurement and four echoes. Blind submission protects the independence that made the panel worth assembling in the first place.
The discussion still happens, and it still matters. But its job changes. It is no longer there to manufacture a verdict. It is there to explain the scores, to surface the disagreement that the numbers already exposed, and to dig into exactly where two careful people saw the same hour differently. That disagreement is the most useful thing in the room, and the old debrief buried it.
The shortlist is ranked by the framework
What you receive is a shortlist ranked by the framework you built, with the scores attached and the divergences visible. Not a single anointed favorite carried out of a room by the strongest voice. A ranking you can interrogate, defend to your board, and revisit later when you want to know whether the screen actually predicted the hire.
None of this removes judgment from hiring. It places judgment where it belongs, in the criteria you set deliberately and the scores each person recorded alone, and out of the part of the process most vulnerable to whoever happens to talk first. The decision is grounded in the framework rather than the strongest voice in the debrief. That is the only claim the method makes, and it is enough.