2 min lesson
Self-score the run
Talk this through in your own words: "You're self-scoring a recorded mock deep-dive. Which signal best predicts you'll pass the real round?" Finish with the next move.
Step 1 of 2
Self-score the run
- Defended with evidence
- Every claim was backed by a number, a baseline or a measurement, not by tone.
- Named rejected alternatives
- For the key forks you said what you didn't do and why - owner, not passenger.
- Volunteered a real limitation
- You surfaced a genuine weakness before any question forced it out.
Cursor screens hard for caring more about what's true than about being right. In the deep-dive that means a volunteered limitation scores higher than a defended flaw. If a reviewer finds a hole you knew about and hid, you fail the value, not just the question.
Pre-bake your weakest-part answer before you ever present. Decide the one limitation you will volunteer and rehearse stating it plainly with the next experiment you'd run. Candidates who improvise this under pressure either understate it (reads as hiding) or spiral (reads as no confidence). A rehearsed, honest weakness reads as a scientist who already pressure-tested their own work.
A flashy team result you contributed one slice to collapses on the third “why,” because you can't defend decisions you didn't make. Pick the project where you chose the optimizer, designed the eval and read the noisy metrics yourself, even if it's smaller. You can answer every follow-up on something you actually drove.