hiring-loop-design ยท diff
git:20260718.7c1285c to git:20260728.fac4544
8 added, 7 removed. Audit A to A.
---
name: hiring-loop-design
description: Design a hiring loop where each interview gathers distinct signal against an anchored rubric and the debrief resists groupthink. Use when standing up or fixing an interview panel and you want a decision based on evidence, not overlapping impressions.
---
# Hiring loop design
An interview loop is a measurement instrument, and most are badly calibrated:
four interviewers ask variants of the same coding question, then average their
gut feelings in a debrief steered by whoever speaks first. Structured loops
predict on-the-job performance far better than unstructured ones. The design
work is deciding what each hour measures and holding the panel to it.
## Method
1. **Map competencies to interviews, one owner each.** List the competencies the
role actually needs (coding, system design, debugging, collaboration, domain
depth) and assign each to a specific interviewer. When two slots both probe
algorithms, you buy redundant signal and leave a competency untested.
2. **Standardize the questions and the rubric.** Use a shared question bank with
a scoring rubric that carries anchored examples: what a weak, solid, and
- strong answer to this exact question looks like. Consistent questions are what
- make candidates comparable and what gives the interview predictive validity.
+ strong answer to this exact question looks like. Consistent questions are
+ what make candidates comparable and what gives the interview predictive
+ validity.
3. **Calibrate interviewers to the bar.** New interviewers shadow and
reverse-shadow before they score alone, and the panel agrees what the target
level's "strong" looks like. An uncalibrated interviewer measures their own
standards, and the loop is only as consistent as its softest grader.
4. **Require written feedback before the debrief.** Each interviewer submits a
hire or no-hire call with evidence and a rationale before anyone talks.
Independent judgments logged first are the defense against anchoring, where
the room drifts toward the first or loudest opinion.
5. **Run the debrief on evidence, not averages.** Read the written packets, dig
into disagreements rather than splitting the difference, and weight signal by
how directly the interview tested it. One well-evidenced no-hire on a core
competency outranks three warm "seemed nice" impressions.
6. **Add an independent bar-raiser for consistency.** Include a panelist from
- outside the hiring team whose job is the long-term bar, not filling this seat.
- Amazon's Bar Raiser is the archetype: it counters a hiring manager's urgency
- to say yes just to close an open req.
+ outside the hiring team whose job is the long-term bar, not filling this
+ seat. Amazon's Bar Raiser is the archetype: it counters a hiring manager's
+ urgency to say yes just to close an open req.
## Litmus tests
- Can you name, for each interview in the loop, the one competency it exists to
measure and how it differs from every other slot?
- - Are written hire and no-hire calls locked in before the debrief, or formed live
- in the room?
+ - Are written hire and no-hire calls locked in before the debrief, or formed
+ live in the room?
- Would two different interviewers score the same candidate answer within one
rubric band?
## Boundaries
A loop design decides how to gather and combine signal, not who to hire in a
given case, which still needs human judgment on the assembled evidence. Legal
limits on interview content, level rubrics, and titles like Bar Raiser are
company and jurisdiction specific, so follow local policy and your recruiting
team's process. Calibrating the interviewers' own standards over time overlaps
with the perf-calibration skill.