A process that varies on purpose
The firm's own framing is that interviews are tailored - every interview is somewhat different, because it aims to make the experience relevant to your background and to the specific roles being considered. That is not evasion, and it has a practical consequence: there is no fixed script to rehearse, so depth in what you actually know beats coverage of what you might be asked.
What candidates report
Hedged as candidate reporting rather than firm publication: a recruiter screen covering background and research interests; a technical stage mixing probability, statistics and a data-analysis exercise, sometimes on a supplied dataset; a longer conversation with a working researcher or engineer; and a final round weighted toward fit and team matching.
The consistent theme in those reports is that the style is collaborative rather than adversarial - interviewers offer hints and redirect a candidate heading down a dead end. Take that at face value. A hint offered and ignored reads far worse than a wrong turn corrected.
What to prepare
Hypothesis-driven thinking, demonstrably. Be able to describe a piece of your own analysis as a question, a method, a result, and an honest account of what it did not establish. This is the single highest-yield preparation for a research-oriented process and the one candidates most often skip.
Statistics beyond the formulas. Expect to justify a choice of estimator or a modelling decision, not only to compute.
Working with data, not just about it. A small analysis exercise rewards practical fluency - loading something messy, checking it, and saying what you can and cannot conclude.
Accept the hints. If the style is collaborative, treating it as adversarial is a self-inflicted wound.
Practise the same formats
The question sets below are ours, written in the formats this process is reported to use. They are not Two Sigma's questions, and nothing here reproduces any part of its assessment.