The problem
A few years ago I pulled the numbers on a hiring loop I'd been running for two quarters: 40-some candidates, five-stage loops, four different interviewers rotating through. Same rubric, same take-home, same debug exercise. When I sorted the composite scores by which slot a candidate landed in — first interview of the day versus fourth — the fourth slot averaged almost half a point lower on our 4-point scale. Same interviewers. Same difficulty exercise, randomly assigned. The only variable that moved was what number interview it was for the person holding the clipboard.
Nobody on the team believed me at first, because it doesn't feel like something you'd do on purpose. And you wouldn't — that's the point.
Why it happens
This isn't really about interviewing. It's about sequential evaluation under cognitive load, and it's been documented outside of tech for a while — the best-known version is a 2011 study on Israeli parole boards, where judges granted parole in roughly 65% of cases right after a food break, and that rate drifted down toward near zero by the time the next break rolled around. Same cases, same law, same judges. The only thing tracking with the outcome was how long it had been since the last rest.
Interview loops run the same machinery. Every candidate forces a fresh context switch: new resume, new background, new code, a new person's nervous energy in the room. By the third or fourth switch in a day, most interviewers aren't consciously grading worse — they're unconsciously grading faster. The structured rubric gets skimmed instead of applied line by line. Borderline calls that would've gotten benefit of the doubt at 9 AM get a reflexive "meh, no" at 3 PM because the mental energy to build a real case for "yes" is gone. It shows up as harsher scoring, but the real cause is depleted evaluation effort, not a genuinely worse candidate pool later in the day.
There's a second, quieter effect stacked on top: anchoring. Interviewer four remembers interviewer three's candidate, who was strong, and grades relative to that recent memory instead of the absolute bar. A perfectly good candidate can read as "fine, I guess" purely because they were unlucky enough to follow someone great.
What to do about it
None of the fixes require new tooling, just changing how loops get scheduled and scored.
Cap interviews per person per day. We went from "however many fit on the calendar" to a hard limit of two technical interviews per interviewer per day. The scores stopped drifting within about three weeks of changing this.
Score immediately, before the next context switch. Interviewers write their score and two supporting sentences within five minutes of the candidate leaving the room — not at the end of the day, not "when I get a chance." Delayed scoring compounds the fatigue effect because you're now recalling a faded impression instead of a fresh one.
Track score against slot position, not just against candidate. Once a quarter, pull the data: average score by interview-of-the-day number, across interviewers. If slot four is reliably scoring below slot one by more than noise, that's a calibration problem, not a talent-pool problem, and it's fixable by scheduling alone.
Rotate who goes first. If the same senior interviewer always opens the loop and the same junior interviewer always closes it, you've baked a second bias directly into the schedule on top of the fatigue effect. Rotate the order across the week.
Key takeaways
- Interview scores drift downward across a single evaluator's day even when candidate quality doesn't — this is a well-documented pattern in sequential judgment tasks, not unique to hiring.
- The mechanism is depleted evaluation effort and anchoring on the immediately preceding candidate, not a real signal about candidate quality.
- Capping interviews per interviewer per day and scoring immediately after each one are the two highest-leverage fixes.
- If you've never checked whether your loop's average score correlates with slot position, it's worth one afternoon with a spreadsheet before you trust your last two quarters of hiring data.
Top comments (0)