playbook

Turn five interviewers into one fair verdict

Synthesize a scattered panel into one balanced read — shared strengths, shared concerns, the real disagreements kept visible, and 'culture fit' language flagged as bias, not evidence.

medium ~30 min

when to reach for this

Five people interviewed the same candidate and their notes are all over the place — one wrote three paragraphs, another wrote "seems fine". The hiring huddle is in twenty minutes and the loudest or longest note tends to win. This system turns the scatter into a fair brief: it weights evidence equally rather than by word count, keeps genuine disagreements visible instead of averaging them into mush, and flags language that's really bias dressed up as a verdict — so the panel decides on the job, not on vibes.

gather this first

The five sets of interview notes, each labeled by interviewer and the area they were assigned (e.g. notes-technical.md, notes-values.md) — in a workspace your company has approved for people data.
The role's scorecard or the must-have skills from the job description, so "strength" and "concern" are measured against what the job actually needs.
The interview plan: who was supposed to assess what, so you can tell a real signal from someone straying off their lane.

the workflow

01
Prime Claude on the job, not just the notes

Before any synthesis, ground Claude in what the role requires. A verdict is only fair if "strength" means "strong at the things this job needs" — otherwise you're summarizing opinions, not evidence.
you ask
```
Here is the scorecard for this role and what each of the five interviewers was assigned to assess. Read them back to me in 3 bullets: the must-have skills, who covered which area, and what 'a strong candidate' looks like for this specific role. Don't read the candidate notes yet.
```
what you get back A short read-back confirming the must-haves and the assessment map ("Priya covered system design, Marco covered collaboration…") so the synthesis is anchored to the role, not to whoever wrote the most.

Anchoring to the scorecard first is what keeps the next step from rewarding the longest note instead of the strongest evidence.
02
Synthesize with evidence weighted equally

Now feed all five notes and ask for a balanced read — explicitly telling Claude not to let length stand in for strength of signal, because a one-line "hire" is worth the same as a three-paragraph one.
you ask
```
Now read all five interview notes for this candidate. Summarize: the strengths every interviewer independently touched on, the concerns that recur across notes, and how the candidate maps to each must-have skill. Weight the evidence equally — do NOT give the longest notes more influence, and don't let one detailed reviewer drown out four brief ones.
```
what you get back A balanced brief — e.g. "3 strengths corroborated across notes (system design, ownership, clear comms), 2 recurring concerns (limited mentoring experience, shaky on scaling)" — with each tied to specific notes rather than to whoever wrote most.
03
Surface the real disagreements — don't average them

An averaged verdict hides the exact thing the huddle needs to resolve. Pull the genuine splits into the open, because where the panel disagrees is where the decision actually lives.
you ask
```
Where do the interviewers actually DISAGREE? Don't smooth it into a middle-ground rating — name each split, quote both sides, and say what evidence would resolve it. Separate genuine disagreement from interviewers just covering different areas.
```
what you get back A short list of true splits — "Two interviewers rated communication strong; one flagged it as a real gap; the difference is whether 'direct' read as clear or blunt" — each with both sides quoted and a concrete tiebreaker, not a blended score.

This is the heart of the system. A fair process makes disagreement visible and resolvable, not invisible.
04
Flag bias and 'culture fit' language as its own pass

Run a dedicated check for comments that aren't job-related evidence. Personality, background, accent, and "culture fit" remarks are where bias hides, and naming them lets the panel set them aside on purpose.
you ask
```
Now scan all five notes for anything that's bias rather than job-related evidence: 'culture fit', 'not like us', comments on personality, accent, background, age, parental status, or how someone looked or dressed. For each, quote the phrase, say why it isn't job-related, and suggest the job-related question it should have been instead. Flag only — don't rewrite anyone's verdict.
```
what you get back A short flag list — "'wasn't really a culture fit' → not job-related; the underlying concern may be collaboration, which the panel should test directly" — that the panel can consciously set aside so the decision rests on evidence.

This surfaces language to reconsider; it doesn't overrule a reviewer. You and the panel still make the call.
05
Write the neutral huddle brief

Pull it into a tight read-aloud summary with no recommendation baked in — so the panel walks in informed and the decision stays theirs.
you ask
```
Draft a 6-line brief I can read aloud in the huddle: the balanced read against the scorecard, the recurring concerns, the one genuine disagreement to resolve, the bias flags to set aside, and the single open question. Do NOT recommend hire or no-hire — leave the decision to the panel.
```
what you get back A neutral, read-aloud brief — strengths, concerns, the split to settle, the flags to set aside, and one open question — with no verdict baked in, so the panel decides in the room.

make it your own

**Fewer or more interviewers:** the system is identical for three notes or eight — just be explicit that every interviewer gets equal weight, and ask Claude to call out any must-have skill that nobody actually assessed.
**Reuse the structure:** save the synthesis prompts as a /panel custom command (see the Playbook's *Features* tab) so every panel gets the same fair, bias-checked treatment instead of depending on who runs the huddle.
**Part of the hiring system:** this is the synthesis stage of the *Run a hire from open to offer* capstone — it consumes the structured scorecards that the interview kit produces, which is what makes the equal-weighting possible.

watch out for

Never paste candidate notes into a tool your company hasn't approved for people data — confidentiality comes first, always. Keep names and contact details out of any prompt you're unsure about.
A bias scan catches common patterns, not all bias, and it can miss context only the panel knows. Treat its flags as prompts to discuss, not as a verdict on a colleague's judgment.
Claude synthesizes and surfaces; it never decides. A human always makes the final hire, and the panel owns resolving the disagreements Claude made visible.

you'll end up with A fair, bias-checked huddle brief in about thirty minutes — shared strengths and concerns weighted by evidence not word count, the real disagreements kept visible and resolvable, and the hire decision left squarely with the panel.

Prime Claude on the job, not just the notes

Synthesize with evidence weighted equally

Surface the real disagreements — don't average them

Flag bias and 'culture fit' language as its own pass

Write the neutral huddle brief