Interviews
Score each candidate 0–100. Feedback is shared with admins for the cut decision. Saving again updates your earlier entry.
Code-reading rubric — run this live, screen-shared, on a real (legacy) repo
Take-homes can be relayed to an AI. This interview can't. Drop the candidate into an unfamiliar codebase cold and watch them work. You're testing whether they can read and reason about code, not whether they can recall facts. Score holistically 0–100; a candidate who can only relay AI answers falls apart on dimensions 1, 2, and 4.
| Dimension | What you're probing | Try asking |
|---|---|---|
| 1. Navigation | Finds things in an unfamiliar codebase without hand-holding. | “Open the repo cold. Where does the order total get calculated? Show me.” |
| 2. Comprehension | Understands what code does and why, in their own words. | “Walk me through this method / stored proc. What's it for? Where does its input come from?” |
| 3. Change-making | Makes a small, correct change live. | “Add this validation / fix this off-by-one. Do it now while I watch.” |
| 4. Consequence reasoning | Reasons about blast radius — the legacy-code skill. | “What breaks if we change this column / signature? Who else calls this?” |
| 5. Reasoning under follow-ups | Thinks for themselves; doesn't stall the moment AI is unavailable. | “Why did you do it that way? What if the input is null? Convince me this is safe.” |
Scoring guide: 80–100 navigates + explains + changes confidently and reasons about consequences · 50–79 reads with prompting, shaky on blast radius · below 50 can't navigate or explain the code unaided (the relay signal). Put concrete moments in the feedback box — it's the evidence behind the cut.