Name What Each Round Measures, Then Delete Everything Else

What is each round supposed to establish?

One thing per round, written down before the loop runs. A phone screen establishes whether the basic requirements are real. A technical round establishes whether the person can do the central task. A manager round establishes how they handle disagreement about work. A round that cannot be described in a sentence like that is a habit with a calendar invite.

Then compare that sentence with what interviewers actually write down, because the notes are the part of a round anybody can still inspect afterwards. Text-mining post-interview notes on 7,650 candidates hired at one large Chinese technology company found that the number of job-related capabilities an interviewer named in the notes tracked later performance and promotions, and ran against turnover, at roughly a 2 percent rise in performance per standard deviation of matching between notes and the job analysis ⁴. That is one firm in one country, correlational, and observable only for people who were hired, so read it as a reason to name capabilities rather than as an effect size to plan around.

Why does a separate track leave the front door alone?

Because a side door only reaches candidates who identify into it, which reintroduces the disclosure penalty the programme was built to avoid. The larger cost is what its existence licenses. Once a special track exists, the standard loop stops being anybody's problem, and it goes on rewarding fast improvisation, easy small talk and steady eye contact for everyone else who applies.

The penalty is measured. A correspondence study that sent applications to 6,016 advertised accounting positions, a third of the cover letters disclosing Asperger's Syndrome and a third a spinal cord injury, found the applicants who disclosed received 26 percent fewer expressions of employer interest, with little difference between the two conditions ⁵. Read the limits with it: accounting roles in the United States, a working-paper figure, expressions of employer interest as the outcome, and disclosure written into a cover letter, which is a different act from opting into a hiring track. It is still enough to explain why a disclosure route makes a poor default.

Small talk is not a neutral warm-up. In structured mock interviews with 189 students, the interviewer's overall impression formed during the rapport-building conversation, before any structured question, correlated .42 with that same interviewer's later structured score, and .25 when a different interviewer supplied the score ². Take the second number as the honest one, and take the paper's own caution with it: the effect ran through rated competence rather than through liking, and a correlation is not evidence that anybody decides in the first three minutes. It is still a channel, and it is the channel that penalises somebody who is bad at unfamiliar small talk and good at the work. The same person reading very differently in two formats is the case in the remote screen and the onsite gap.

Whatever the loop calls fit is exposed to the same problem. In a 2024 complaint to the Federal Trade Commission about how one vendor marketed its assessments, the ACLU argued that personality tools measuring traits such as positivity, emotional awareness and liveliness are not clearly job related for most jobs and are directly linked with core aspects of the medical understanding of autism and of conditions such as depression and anxiety ¹. That is an advocacy organisation's contention about marketing claims. No agency has made a finding on it, and the vendor disputes the characterisation. The mechanism it names is the part to take seriously: a construct can be clean on race and sex and still act as a proxy for a disability.

Cut the criteria that are not the job

Take one round's scorecard and cross off every line that is not the sentence you wrote for it. In most loops that removes three things nobody would defend in writing: speed of improvisation, comfort with a stranger, and fluent thinking aloud while being watched. Deleting them costs nothing, because none of the three is what you wrote down for that round, and it reaches candidates who never told you anything about themselves.

What replaces them is mostly subtraction:

• Separate whether the reasoning was right from whether it arrived quickly. Give each of those its own line on the scorecard.
• Allow a candidate to take a minute before answering, and stop reading the silence as a gap. Say so out loud at the start of the round.
• Allow notes and reference material, unless working without either is the job.
• Stop scoring demeanour. Eye contact, energy, warmth and polish are not on the sentence you wrote, and where a tool produces those adjectives from a transcript the underlying error is not evenly distributed either, which is whose words the transcript gets wrong.
• Put the weight on the follow-up. What somebody does when pressed on a claim is closer to the work than how smoothly they opened.

Most of the signal moved to the follow-up anyway, and the follow-up questions that expose whether someone understands their own answer is that version with the questions written out.

There is a reason the unstructured parts go first. Highhouse's review reports that interrater reliability on the traditional unstructured interview is so low that even with a perfectly reliable and valid criterion, interview-based judgments could never account for more than 10% of the variance in job performance, and reports alongside it a 1996 survey of 201 HR executives who rated the unstructured interview more effective than any paper-and-pencil assessment procedure ³. Both numbers are that review restating other people's work: the 10% is a ceiling implied by reliability rather than a measured validity, and the survey records what practitioners believed at the time. The distance between the two is still the thing worth sitting with.

Send the questions in advance and mean it

Send the actual questions, not a theme, at least a day ahead, to every candidate. This is the single change with the best ratio of effort to effect, and the objection to it deserves a plain answer: yes, sending them means you stop measuring surprise. That is the point. Surprise was never on the sentence you wrote for the round.

The usual counter is that prepared answers carry less information. They do, for the first answer. They do not for the second, because a prepared answer is easy to write and hard to defend, and the round should be built to press on it. Doing that consistently across a panel is the mechanics in sending interview questions in advance.

Two more that cost an email each. Tell candidates the format before the round: how many people, how long, whether anything is live, whether a camera is expected. And tell them what the round is for, using the sentence you already wrote, because somebody who knows what is being established can aim at it while somebody guessing is being measured on guessing.

None of this requires a disclosure, a diagnosis, a programme or a budget line, which is the argument for it. In the United States that is also the shape the law expects: 42 U.S.C. 12112(d)(2)(A), enacted in 1990, bars an employer from asking a job applicant whether they have a disability, or about the nature or severity of one, before a conditional offer, while allowing questions about the ability to perform job-related functions ⁶. That governs what an employer may ask. It does not bar telling applicants what a round involves and inviting requests, and the wording you use is worth checking with counsel. A candidate who would have asked for questions in advance gets them without asking. A candidate who would never have asked gets them too. And the interviewer ends up with a round that measures the thing they wrote down, plus notes naming the capabilities that were actually demonstrated, which is the only part of an interview anybody can still argue with six months later.

See how it works >