You are chairing the debrief for a senior candidate. Two interviewers who saw the same person are firm and opposite: one strong hire, one no hire. How do you run the room and how do you decide?

Whoever speaks first has already decided the meeting

The single largest determinant of a debrief's outcome is the order of speaking, and a manager who opens with "so, what did everyone think?" has handed the decision to whoever is most confident or most senior. Anchoring in a hiring debrief is not a subtle effect; a definite verdict in the first thirty seconds visibly moves everyone who was uncertain, and the people who were uncertain are usually the majority.

So the mechanics come before the discussion. Written feedback and a score against each competency, submitted independently, before the meeting opens and with no visibility of other submissions. Then in the room, read the evidence before the verdicts: ask each interviewer to describe what the candidate did and said, not whether they liked it. Take the least senior interviewer first and yourself last. And if the written feedback did not arrive, the correct move is to postpone rather than run the debrief, because a verbal-only debrief is a popularity contest with a paper trail.

The reason to be strict about this is not fairness in the abstract. It is that a split panel is the only situation where the process actually earns its cost, and it is exactly the situation where an unstructured room produces a decision nobody can reconstruct a month later.

Two disagreements that look identical and are not

Once you have the evidence in the open, the split resolves into one of two things, and confusing them is the most common failure.

The first is a disagreement about what happened. One interviewer saw a candidate who reasoned carefully towards a good design; the other saw someone who needed three prompts to get there. Both may be accurate about different forty-five minutes. Interrogate the artefacts rather than the impressions: what was the question, what did the candidate produce, what did the interviewer have to supply. Very often the design interviewer asked a much harder question than they realised, or asked about a system the candidate had genuinely never encountered, and the "no" is a fact about the question. This kind of split is resolvable in the room, and the resolution is frequently that one interviewer's evidence is thinner than their conviction.

The second is a disagreement about the bar. Both describe the same behaviour and one thinks it clears senior while the other does not. This is not resolvable by more discussion, because there is nothing left to establish about the candidate. It is a calibration problem, and the only honest handling is to go back to the written expectations for the level and ask which specific expectation the evidence meets or fails. If your organisation has no written expectations for the level, the split is your fault rather than theirs, and saying so in an interview shows more than defending the debrief would.

What the chair says	Why it works
"Before opinions — Sam, what was the question and what did they produce?"	Puts evidence in the room before any verdict can anchor it
"You both watched the same system design. Where do your accounts differ?"	Forces the distinction between disputed evidence and a disputed bar
"Which line of the senior expectations does that fail?"	Converts a feeling into something checkable, or exposes that there is nothing to check
"What would you have needed to see to change your mind?"	Reveals an unfalsifiable objection, which is the tell for a bias rather than an assessment
"I am hearing a bar disagreement, not an evidence disagreement, so we are not going to settle it tonight."	Stops the room grinding towards fatigue-consensus

The decision has an owner

A debrief is not a vote and pretending it is one produces the worst outcome available: the room negotiates towards the middle, someone proposes a fifth interview, and the candidate waits ten days for a process that has stopped generating information. Additional rounds after a split are usually an attempt to avoid making a decision, and they rarely reverse anything except the candidate's interest.

State the model plainly at the start of the meeting: the panel supplies evidence and assessment, and the hiring manager decides. Then decide in the room. If the evidence genuinely does not cover the competency in dispute — not "we are unsure", but "nobody assessed this" — one targeted follow-up round with a named competency and a named interviewer is legitimate. That is a different thing from another general onsite.

When you decide against a strong voice on the panel, tell them why, in terms of the evidence, and do it before they hear the outcome from anyone else. An interviewer whose no-hire is overridden without explanation stops writing candid feedback, and you have just degraded the instrument you rely on.

The bias operating on you, not on them

The candidate you are least equipped to assess is the one you need. An open requisition for two quarters, a team carrying an unowned service, and a recruiter reporting a thin pipeline all push in the same direction, and none of them appear in anybody's written feedback. This is where an interviewer probing you will push, because the honest answer is uncomfortable: the reason marginal candidates get hired is almost never that the panel was fooled, it is that the manager wanted it resolved.

Two guards are worth naming. First, decide before the debrief what you would do if the answer is no — which work you will not staff, which date moves — so the cost of no is concrete rather than vague dread. Second, treat the ambiguous senior hire as the expensive case it is: the cost of a wrong senior hire is not the salary, it is a quarter of the team's attention, the design decisions taken in the meantime, and the exit process, against a vacancy that costs you delivery you can plan around.

The mirror-image bias deserves the same scepticism. An interviewer who has never recommended a hire is not maintaining the bar, they are declining to make a judgement, and their no carries less information than the panel usually gives it.

A split panel is data about your loop

Whatever you decide about the candidate, a firm two-way split means something about the process, and the strongest answers reach for that unprompted. Look at whether the two interviewers were assessing the same competency at all — a loop where two rounds nominally cover "technical depth" and one is a coding exercise while the other is architecture will produce contradictory scores forever. Look at whether the questions are comparable between interviewers, since two people asking their own favourite problem are running two different bars by construction. And look at whether the interviewer whose evidence was thinnest has been calibrated recently, or ever.

Then do something small and specific: pair the two of them on the next candidate at that level and compare their independent scores afterwards. That produces more improvement than a training session, and it is the answer that tells an interviewer you have actually run a loop rather than read about one.