Interviewer Guide
How to run an interview using this bank — a methodology, not a question list.
Before You Start
- Pick questions from
QUESTION_INDEX.md matched to role and level — see PRACTICE_TRACKS.md.
- Read each question’s full entry beforehand, including “Common Mistake” and “Interviewer Follow-Up” — the follow-up is where you learn the most.
- Decide time budget per question before starting.
Running a Session
- State the scenario, then stop talking. How a candidate clarifies an underspecified problem is itself signal.
- Let them ask clarifying questions. Reward the instinct to check nullability/edge cases before writing SQL.
- Use the Interviewer Follow-Up when they finish. This is where shallow-but-correct and deep understanding become distinguishable.
- For Debug/Optimize/Incident questions, watch the process, not just the answer. Guessing “add an index” immediately, without asking for EXPLAIN first, is a weaker signal than asking for more information first.
Probing Technique
- If they give the “Common Mistake” answer, ask them to predict what the query actually returns rather than just saying “that’s wrong.”
- If solved quickly, move to the Senior-Level Extension instead of a new question — depth on one topic tells you more than breadth across shallow ones.
- Silence after asking is normal; resist filling it.
Scoring
Use SCORING_RUBRIC.md’s eight dimensions per question, not one pass/fail. A candidate can score well on correctness and poorly on performance awareness on the SAME question.
Red Flags
- Cannot explain why their own query works, at any level
- Every performance question answered “add an index” regardless of what EXPLAIN shows
- No tradeoff articulated on any Level 4+ question
- Cannot continue after one wrong answer
Strong Signals
- Clarifying question before writing SQL when genuinely ambiguous
- Proposes verifying a hypothesis before committing to a fix
- Articulates a real tradeoff unprompted on Level 4+ questions
- Recovers gracefully from an incorrect first attempt
Avoiding Unfair Questions
- Don’t require memorized exact syntax under time pressure — open reference material is fair; reasoning is what’s tested
- Credit valid alternative approaches not listed in “Alternative Solution”
- Calibrate difficulty to the role — Level 0/1 is the appropriate bar for many roles, not “too easy to ask”
Evaluating Reasoning, Not Memorization
The biggest failure mode: rewarding memorized LeetCode-style patterns without checking if the candidate can explain WHY they work. Every “Explanation” field exists so you can ask “why” as a follow-up to any correct answer.