Seniority Shows Up in Decisions Under Ambiguity, Not in Titles

What actually separates a senior candidate from a mid-level one?

Level shows in the biggest call they were trusted to make with no safety net, and in what it cost when that call was wrong. That is not on a resume, and years carry less than they look: prehire work experience correlates .06 with later job performance, and task-relevant experience does no better ¹. That is a finding about performance, and level is a separate question. Two people with the same title and the same tenure can sit two levels apart.

The standard proxy list does not survive contact with the current market. Reporting line, team size and budget authority are all claims inside a document the candidate controls, and the tell that used to make an inflated claim visible, a sentence that read a little too large for the rest of the page, is gone now that drafting help is universal. The list also misses the thing it is trying to reach. A person can manage a team of twelve and never make a call that could not be reversed by their manager on the same day.

What does separate levels is decision scope under ambiguity. At mid-level, the question is usually available somewhere: an approach exists, a precedent exists, somebody senior will confirm it. At senior, the answer does not exist yet, the tradeoff has a real loser, and somebody has to own the consequence. That is why the most informative question in a screen is about a decision that went wrong. A candidate who owned real scope can price their own mistakes; a candidate who did not will describe a mistake somebody else made.

One stage earlier, the same reading problem arrives as a stack of applications that all look equally polished. What to actually screen on when every resume looks perfect is the top-of-funnel version of the question this article answers in the room.

Ask for three decisions, then keep asking

Three questions, in this order: a decision you owned end to end, a decision you escalated and why, and a decision that went badly and what it cost. Ask them of every candidate for the role, in the same words, and rate the answers on the same scale. The structure does more work here than the wording of any one question.

The numbers back that up. In the 2022 re-analysis of the selection literature, structured interviews estimate at .42 and unstructured interviews at .19, and before any correction the raw observed figures were .32 and .13 ². Structured there is a coding of the interview format: the same questions, rated on a common scale. The US Office of Personnel Management adds a third element to that definition, which is that the interviewers agree in advance what an acceptable answer looks like ³. Structure is a property of how the interview is run, not a script somebody buys.

The follow-ups carry the signal, so plan two per question and do not skip them when the first answer sounds good:

• Who disagreed with you, and what did they say? A real decision had an opponent. An invented one rarely does.
• What did you give up? Every senior call has a loser. A candidate who cannot name it was describing an outcome.
• What would have changed your mind at the time? This separates judgment from hindsight, which is the part that survives preparation.
• What did it cost, in a number or a date? Vagueness here is the most reliable tell in the whole set.

A thin answer to the third question sounds like this: the project shipped late and I learned a lot about scoping. A real one sounds like this: I chose to hold the release for the migration, it slipped eleven days, the sales team lost two renewals in the quarter, and I would make the same call again, because rolling back mid-migration would have cost the data. The second answer is checkable, priced and owned. That is the level.

Write the level definition before the screen, in behaviors

Write two or three sentences per level describing what a person at that level decides without asking, before anybody talks to a candidate. Levels written in years or in headcount cannot be applied to an answer. Levels written in decisions can. Without them, two interviewers hear the same story and file it at different levels, and the debrief turns into a negotiation between confident people.

The evidence on how those readings get combined is blunter than the evidence on how they get gathered. In a meta-analysis of selection and admissions studies, applicant data combined by a stated rule correlated .44 with job performance, against .28 when the same kinds of data were combined by expert judgment, and the authors classify a group consensus meeting as the second kind: what makes a method holistic is that the pieces get combined by judgment rather than by something applied the same way each time ⁴. That comparison rests on nine studies for job performance. Keep the debrief, which is how evidence gets surfaced and corrected, and agree the combination rule before it starts.

What gets written down after the interview matters too. Text-mining the post-interview notes on 7,650 candidates hired at one large Chinese technology company, researchers found that the number of job-related capabilities an interviewer named in their notes was positively related to later performance and promotions and negatively related to turnover, with a one standard deviation rise in the note-to-job-analysis match corresponding to roughly a 2 percent rise in performance ⁵. That is correlational, at one firm in one country, and only candidates who passed the interview could be observed. What it supports is notes that name the capabilities the role calls for. It says nothing about whether taking notes at all improves a decision.

The practical version fits on one page. For each level, name what the person decides alone, what they escalate, and what the largest reversible and irreversible calls look like. Then have every interviewer mark which of those they saw evidence for, in the candidate's own examples, before the debrief starts.

Verify the two claims that live outside the room

Scope and reporting line are the two claims a former manager can settle from memory, so put both of them on a reference call. Ask what the candidate owned without needing approval, and how large the biggest call was that they made alone. Use the same wording for every reference on every candidate, and keep notes you can compare side by side.

The validity figure usually quoted for reference checks, .26, comes from the 1998 table, and reference checking was one of eight procedures excluded from the 2022 re-analysis for insufficient information, which makes it at once the only published number and the least checkable one in the set ⁶. The US Office of Personnel Management's own guidance says reference checks predict better than years of education or job experience but less well than a cognitive ability test, that adding structure improves them, and that written requests produce low response rates and thin information; it budgets about 20 minutes per contact by phone with a minimum of three contacts ⁷.

So a reference call is a verification instrument, not an evaluation instrument. Scope and reporting line are things somebody else watched happen, which is why a call can settle them. The phrase they were great is reaching for something no call settles. Getting past that phrase is its own skill, worked through in how to reference-check for AI judgment when the reference just says they were great.

What is left after all this is a level you can defend in a sentence: this candidate has owned decisions of this size, with this consequence, confirmed by a person who watched them do it. That sentence beats a title, and it beats a number of years, and it is the one thing in the file that a model did not draft.

See a sample report >