§ CAPABILITY

Technical Interview Loop Design

A loop where each stage measures something different and the stages combine into a decision. Structured, because unstructured interviews mostly measure how much the interviewer enjoyed the conversation.

Who
Founder holds the engineering leadership seat; the delivery team executes underneath it.
Request a call How the work runs

What you're seeing

Three of the four interviewers ask about the same thing.
Usually means Nobody assigned a purpose per stage, so each interviewer defaulted to what they are comfortable asking. The loop is longer than it needs to be and no better calibrated for it.
The decision meeting is settled by whoever speaks first.
Usually means Assessments are being formed in the room rather than before it. The first confident opinion anchors everyone else, and the loop's output is now one person's judgement wearing four people's names.
Candidates who employ themselves elsewhere drop out at the exercise stage.
Usually means The take-home is unbounded in practice whatever the stated cap says. It is selecting for available evenings, which correlates with nothing you are hiring for.
Two people were interviewed for the same role and cannot be compared.
Usually means The stages differed, or the criteria did. Without a common frame the decision defaults to who advocated hardest, and six months later there is no record of why.

One thing per stage

When two stages measure the same attribute, the loop is longer than it needs to be and no better calibrated for it.

So the design starts from the role rather than from the interviews you already have: what three or four things does this loop have to establish before an offer is defensible. For a senior backend role that might be system design under real constraints, code quality on unfamiliar code, debugging under uncertainty, and how they work with someone who disagrees with them.

Each of those gets one stage. Nothing gets two. This usually means retiring a stage that somebody enjoys running, and that conversation is part of the work.

It also forces the role definition to exist, in writing, before the loop is built — which is frequently the actual missing piece rather than the interviews themselves.

Anchors, not adjectives

A rubric that says “good communication” is a mood with a form attached.

Two interviewers reading it will score the same candidate differently and both will be correct by their own interpretation, which means the rubric added a number without adding agreement. An anchor is something a person can point to afterwards: explained a trade-off without being asked, changed position when shown a counter-example, asked about failure modes before asking about scale.

Those are observable. They can be disagreed about productively, which is what calibration needs, and they survive the six-month question of why someone was rejected.

Write before you talk

Independent written assessments submitted before the debrief, every time.

The mechanism this defends against is simple and reliable: the first confident opinion in the room becomes the room’s opinion, and seniority makes it worse. People do not intend to defer and mostly do not notice they have. Collecting positions first makes disagreement visible instead of dissolving it, and a loop where four people wrote down four different readings is producing exactly the signal it was built to produce.

Where it sits

This is a capability inside Engineering Hiring, which carries the whole pipeline — sourcing brief, screen, loop, offer and the first ninety days — and this is the stage most companies get wrong first.

It depends on two neighbours to work. Interviewer Calibration is what makes a rubric mean the same thing to different people; without it the loop is structured on paper and unstructured in practice. And Career Ladders is what defines the level a candidate is being assessed against, because a rubric without a level is a rubric without a threshold.

How the work runs

  1. Decide what the loop must establish

    Three or four things, derived from the role definition. A loop that tests everything tests nothing well and takes a candidate two weeks.

  2. Assign one to each stage

    No two stages measuring the same attribute. Duplicate coverage is where candidate time and interviewer time both go.

  3. Write rubrics with observable anchors

    What a strong answer contains, in terms someone can point at afterwards. A rubric that says 'good communication' is a mood.

  4. Design the debrief

    Written assessments submitted before discussion, so the loudest voice does not anchor the room.

What arrives

  • A loop with a stated purpose per stage
  • Rubrics with observable anchors per level
  • A bounded practical exercise reflecting real work
  • A debrief format with independent written assessment first

What it costs your team

Two design sessions plus review from the engineers who will run the loop.

Next step

Tell us where this is breaking and we'll say what it takes to fix.

Request a call

How we decide

  • One thing per stage, derived from the role

    Costs It means dropping a stage somebody likes running, and telling them why.

    Two stages measuring the same attribute cost candidate time and interviewer time and add no signal. Deriving the three or four things the loop must establish from the role definition also forces the role definition to exist, which is frequently the actual missing piece.

  • Rubrics use observable anchors, never adjectives

    Costs Writing them is slow and requires arguing about what a strong answer contains.

    A rubric that says 'good communication' is a mood with a form attached — two interviewers will score the same candidate differently and both will be right by their own reading. An anchor someone can point at afterwards is what makes the assessment comparable, and it is what makes a rejection explainable to the candidate and to yourself later.

  • Written assessments are submitted before the debrief

    Costs It adds a step and slows the decision by a day.

    The first confident opinion in a debrief anchors the room, and seniority amplifies it. Independent written assessments collected first are the cheapest available correction — the disagreement is then visible and discussable, instead of being resolved before it was noticed.

  • The exercise is bounded and graded against the bound

    Costs A capped exercise reveals less than an uncapped one, and some candidates will exceed it regardless.

    An uncapped take-home measures available evenings. Grading against the best submission received re-imposes the same bias even when the cap is stated, so the rubric has to describe what a good two hours looks like rather than what a good submission looks like.

Frequently Asked Questions

One where every candidate for a role faces the same stages, is asked comparable questions, and is assessed against written criteria agreed in advance. It is the single change with the most evidence behind it for improving hiring accuracy, and it costs almost nothing beyond the discipline of doing it.
Pairing on a realistic problem, most of the time. It shows how someone works with another person on unfamiliar code, which is close to the actual job. Whiteboard algorithm rounds largely measure recent interview preparation, and take-homes measure available time unless they are bounded hard.
Four stages and under four hours of candidate time in total. Beyond that you lose people who are currently employed, which is most of the people you want to hire. A loop that takes two weeks of elapsed time loses them to a competing process regardless of length.
Usually. It keeps the decision close to the person accountable for it, and a rejection earlier in the loop has not consumed their time. The exception is a role where the manager is the main thing being sold, in which case an early conversation is worth the cost.
Design for it rather than hoping. A bounded pairing exercise on realistic code, a written component for people who think better in writing, and questions sent in advance where the goal is depth rather than surprise. Interview performance and job performance are different skills, and a loop that does not account for that is measuring the wrong one.

Sources

Page reviewed