Every interviewer holds a different bar

Five sincere people, five different bars

The spread is not incompetence. Each of those bars is a defensible reading of the same job description, held by someone doing their best. What the spread costs is comparability: two candidates who never met the same standard cannot be ranked against each other, and the ranking is the entire output of a hiring loop.

The candidate at nine a.m. and the candidate at six p.m. did not meet the same company.

Calibration helps, and then it decays

The usual fix is calibration: rubric documents, shadow interviews, debrief rituals. These help. They also decay. Rubrics drift the week after they are written, shadowing does not scale past a handful of loops, and a debrief is where five private bars negotiate a settlement, not where a shared one gets applied.

The decay has a mechanism worth naming. A written rubric competes with a remembered one, and the remembered one is free to consult. Under time pressure the free option wins, which means calibration has to be re-paid every loop or it quietly stops being in force.

Make the bar an artifact

Our position is that the bar should be an artifact, not a memory. Written once, in plain language, and then held by something that does not get tired, rushed, or generous at the end of the day. People still decide, reading the reasoning and overruling it when they know better. But the standard itself stops living in anyone's head.

This is the founding argument of Evals, and it is why the product begins with writing rather than with interviewing. If you cannot state the bar, you do not have one. You have interviewers.