Flatten an interview to a number and two very different candidates become identical. One scored a four by racing to a solution and defending it stubbornly; the other scored a four by moving slowly and catching their own mistake. Same number. You want the second one.
What the aggregate hides
This is why every score in Evals ships with the reasoning that produced it, not a summary written after the fact but the criterion-by-criterion account of what the examiner saw and how it weighed it.
The score is the index. The reasoning is the record.
Reading reasoning is a learnable skill
Look for where the examiner hesitated
Criteria scored with caveats are where the interesting judgment lives. A clean five tells you less than a hedged three.
Look at what the candidate did when wrong
No aggregate captures the recovery, and the recovery is most of what you are hiring.
Follow surprise into the transcript
When the reasoning surprises you, that is either the examiner catching something you would have missed, or a rubric flaw you now get to fix.
Either way you learn something. A bare number teaches nothing, twice: once about the candidate, once about the bar.
watchglass (2026). Reading the transcript: what reasoning shows that scores cannot. The notebook, volume № 01, § 03. watchglass.app.