A real one, unedited
This is what lands after a round. Numbers at the top are arithmetic. Everything below them is judgement, and it says so.
Session 04 · Resume Grill
11 Aug 2026 · 12 questions · 9:12 spoken
You know your projects. You describe them backwards — the fix first, the problem never. Two of twelve answers had no stated constraint, which is what an interviewer is actually listening for.
Delivery
Computed in code · not asked of the model- 17
- Filler words
- 168
- Words / minute
- 9:12
- Speaking time
- 46s
- Longest answer
um ×9, like ×5, basically ×3
measured over talk time only
of a 14:40 session
Q7, the migration question
Answer by answer
Judgement · LLM, schema-validatedQ3 · Why Redis in front of Postgres, and what did it cost you?
"We added Redis because the database was slow and it got a lot faster after that."
"Reads outnumbered writes about 40 to 1 and p95 on the menu endpoint was 900ms. We cached the payload for 90 seconds, which took p95 to 60ms and bought us a 90-second staleness window — acceptable because menus change daily, not hourly."
Q8 · Tell me about a time you disagreed with a teammate's approach
"There was some friction but we sorted it out and shipped on time, so it was fine in the end."
"He wanted to hand-roll auth; I wanted a provider. I wrote down the two failure modes I was worried about, we spiked the provider for a day, and the spike settled it. I'd still push for the spike over the argument."
Structure notes
- Nine of twelve answers opened with the solution. State the problem first.
- No numbers volunteered until asked. You have them; lead with them.
- Answers ran long past the point they were made — the interrupt fired twice.
Red flags
- Couldn't name a tradeoff for a decision you made yourself.
- Credited "the team" for work the resume attributes to you.
- Said "I don't remember" about a project listed as current.
Validated before you see it
The judgement half comes back as JSON against a fixed schema. If a field is missing or malformed, the round is re-scored rather than shown to you half-built.
{
"score": 6.5,
"answers": [{ "said", "strong" }],
"structure_notes": [3],
"red_flags": [3]
}
Delivery metrics are not in this payload