Went through the Jane Street DS loop for a research data role in 2026. It's short on documentation so writing this up.
First thing: Jane Street does not have a large traditional DS function. The role I interviewed for was research-adjacent, closer to quant research assistant than a product DS. If you're expecting a standard SQL + A/B test + ML model interview, manage your expectations. The signal they're looking for is sharper.
What the rounds covered:
Statistics and probability: Heavy. I got asked to derive results from scratch, not just describe them. Think: conditional probability chains, Bayesian updating, distributions. One question was a variant of a classic probability puzzle with a twist that required careful reasoning. I got it wrong the first time and had to walk back. The interviewer let me re-approach, which felt intentional.
SQL: Yes, there was SQL, but it was paired with reasoning questions. Not just "write a window function," more like "given this query result, what might the underlying data problem be?" They care about whether you understand what SQL is doing, not just syntax.
Case / business reasoning: The case was more analytical than strategic. Given a dataset with anomalies, what hypotheses would you form, how would you test them, what would you do if your first hypothesis was wrong? Think: structured analysis under uncertainty.
No ML modeling round. No sklearn, no model selection, no feature engineering questions. If you're prepping ML depth for this, you're prepping for the wrong loop.
What I'd do differently: Spend more time on probability theory refreshers, not just SQL drills. The stats depth caught me more than I expected.
Overall the loop was intellectually interesting. Didn't move forward but felt like a fair process.