A loop is well designed when every stage has a distinct question it answers, every competency has exactly one owner, and the whole thing fits inside a candidate-hour budget you would accept if you were the candidate. Most loops fail all three: they repeat the same conversation four times, leave the most important competency unowned, and cost strong candidates so many hours that only the desperate finish.

Start from the decision, not the calendar

Write down the 5–7 competencies that decide success in {{Role title}}, then design backwards: what is the cheapest reliable way to observe each one? Some competencies need an hour of working together; others need one pointed question in a screen. The loop is whatever that backwards pass produces — usually 3–5 stages, 4–6 total candidate hours for senior roles, less for junior ones.

Stage goals that do not overlap

If two stages would answer the same question, delete one. Confirmation is not a stage goal.

Assigning interviewers

• One owner per competency; write the mapping down and put it in the calendar invite. An interviewer without an assignment defaults to re-running their favorite interview.
• Two interviewers per session maximum. Panels of four intimidate candidates and dilute accountability for the notes.
• Mix the loop deliberately: seniority levels, functions, and demographics. Candidates read who you send as a statement about who belongs.
• Only trained interviewers touch live candidates — everyone else shadows first. Interviewing is a skill, not a seniority perk.

Anti-bias mechanics to build in, not bolt on

• Same stages, same order, same primary questions for every candidate in the search. Ad-hoc extra rounds for candidates someone is unsure about are where bias hides — if the loop needs another stage, add it for everyone.
• Scorecards submitted before any discussion; debrief runs junior-first.
• Decide the weighting of stages before interviewing begins, and put standardized skill assessments and work samples early, where they anchor the process in demonstrated ability rather than pedigree.
• Track pass-through rates by stage over time. A stage that filters one group disproportionately is a broken instrument until proven otherwise.

Loop anti-patterns

• The endless loop: six-plus rounds signal a committee that cannot decide. Every added stage costs offer-acceptance rate; the strongest candidates are gone by round five.
• The gauntlet clone: copying a big-company loop wholesale, brainteasers included, for a ten-person team with completely different constraints.
• The unowned veto: someone outside the loop kills a candidate in the hallway. If they want a vote, they join the loop, run the guide, and file a scorecard.
• The stealth extension: {{Just one more chat}} after the debrief, uncalibrated and unscored. It feels diligent; it is noise with veto power.
• The trivia round: puzzle questions unrelated to the work. They measure puzzle practice and confidence, and reliably favor candidates with time to grind.

Instrument the loop

Review quarterly: time-to-decision per stage, pass-through rates, offer-accept rate, interviewer scorecard-completion rate, and — the number nobody tracks — how often each stage's score actually changed the final decision. A stage that never changes decisions is candidate-hours you are spending on ceremony.