Jane Street Interview Questions — Page 3

Jane Street·Software Engineer·Technical Phone Screen

May 2026

Jane Street SWE interview, got a stack machine interpreter problem. Pretty self-contained but the type system angle made it trickier than it looked at first glance.

Implement an interpreter for a simple stack machine that supports PUSH, POP, POP_AND_PRINT, and ADD instructions, where values on the stack are either integers or strings and the ADD operation behaves differently depending on the types of the two operands.

“The ADD rule is where it gets you.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

May 2026

Jane Street Data Scientist interview with a heavy focus on auction theory and probabilistic reasoning. The main problem was a sealed-bid auction for a coin-flip payout, and it got progressively nastier with information asymmetry thrown in. Not a coding round at all, more like a quant brain teaser stretched into a full conversation.

You and one other bidder are in a sealed-bid, first-price auction for a box that pays $1 per head from 200 fair coin flips. Neither of you has seen any flip outcomes. What do you bid and why?
Now your opponent secretly saw 10 of the 200 coin flips before bidding, but you saw nothing. You know they have this extra information. How does your strategy change?
What if your opponent had seen all 200 flips before bidding? What's the most you should ever bid, and what's your expected profit?
How does your analysis change if this were a second-price auction instead of first-price? Does the winner's curse go away?
What's the maximum you'd pay upfront, before the auction starts, for the right to observe 10 coin flips yourself?

“Expected value is $100, standard deviation around $7.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

May 2026

Jane Street Data Scientist interview with a heavy probability/statistics focus. The core problem was a multivariate correlation question that required both a proof and a construction, which is pretty typical for quant-adjacent roles there but still caught me off guard with how much rigor they expected.

Given seven random variables on the same probability space, each with mean 0 and variance 1, identically distributed, and with the same pairwise correlation rho for every pair, what is the minimum possible value of rho? You need to prove the lower bound is tight by showing it can't go lower AND constructing a valid joint distribution that actually achieves it.

“This wrecked me for the first few minutes.”

View Post

Jane Street·Machine Learning Engineer·Onsite - Coding / Algorithms

May 2026

Jane Street ML Engineer interview with a coding round that went deeper into system design territory than I expected. The Connect-N problem sounds like a toy game until they start asking about infinite boards and complexity tradeoffs.

Design and implement a Connect-N game on an infinite board where pieces fall under gravity. Implement an insert function and a check for N-in-a-row across all four directions after each move.
After placing a piece, how do you efficiently check for N-in-a-row without rescanning the entire board?
What is the time complexity per insert, and how would you handle very large values of N efficiently?

“My first instinct was a 2D array and I almost said it out loud before catching myself.”

View Post

Jane Street·Software Engineer·Technical Phone Screen

May 2026

Jane Street SWE interview that leaned harder into spreadsheet mechanics than I expected. The whole thing was basically one extended Excel problem about collateral valuation, which felt oddly niche but made sense in context.

You have an Excel sheet with columns for Asset, Market Value, and Haircut %. Using only cell references (no hardcoded numbers), write formulas to calculate each asset's haircut-adjusted value and the total adjusted collateral value. Walk through your formula choices, how you'd handle absolute vs. relative references, and how you'd make the output update automatically when inputs change.

“This was more involved than it looks.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

May 2026

Jane Street Data Scientist interview that leaned heavily into probability theory. The core problem was a two-day rain scenario where you had to derive bounds on union probability without assuming independence, then reason about real-world weather dependence. Solid follow-ups pushed further into three-day extensions and conditional inference.

Given that rain on Saturday has probability p and rain on Sunday has probability q, with no assumption about independence between the two days, what is the full range of possible values for the probability that it rains on at least one of the two days? Show that both endpoints are actually achievable.
If rain events typically last only one day, where in that range does the true probability of at least one rainy day fall? What if rain events typically persist for two or more days?
If you are additionally told that both days are rainy with probability r, what is P(at least one rainy day) exactly, and what values of r are even feasible given p and q?
If the two days were exactly independent, where does P(at least one rainy day) fall within the range you derived? Can it ever touch either endpoint when both p and q are strictly between 0 and 1?
Extend the problem to a three-day weekend where each day has its own marginal rain probability. What is the range of possible values for the probability that it rains on at least one of the three days?
Given only P(A union B), p, and q, what can you infer about P(A intersect B) and about the conditional probability P(B given A)?

“This is where inclusion-exclusion saves you.”

View Post

Jane Street·Software Engineer·Technical Phone Screen

Jun 2026

Jane Street software engineer interview with a coding question on evaluating a tokenized infix arithmetic expression. The problem was well-scoped but the precedence and parentheses handling had enough edge cases to keep things interesting. No fluff, just straight into the technical stuff.

Given an already-tokenized arithmetic expression as an array of strings (integers, operators +/-/*, /, and parentheses), evaluate it and return the integer result. Standard operator precedence and left-to-right associativity apply, and all divisions are guaranteed exact.

“The 'already tokenized' part is doing a lot of work in this problem.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

Jun 2026

Jane Street data scientist interview, all probability puzzles, no coding, no case studies. The interviewer just handed me five problems and expected me to think out loud the whole time. Felt more like a math exam than a job interview, which I was not fully prepared for.

You eat pizza on Saturday with probability 0.4 and on Sunday with probability 0.3. No information is given about how the two days relate. What is the range of possible values for the probability of eating pizza at least once over the weekend? Describe a dependence structure that achieves each extreme.
You flip 250 gold coins and 250 silver coins. Gold heads score 3 points, silver heads score 1 point, tails score 0. Given that the total score is exactly 900, which combination of gold heads and silver heads is most likely?
An ant starts at one vertex of a unit cube and moves each second to a uniformly random adjacent vertex. What is the expected number of steps to reach the vertex diagonally opposite the starting point?
You have N shoelaces in a pile with 2N free ends. Repeatedly pick two free ends at random and tie them together until no free ends remain. What is the expected number of closed loops formed?
Two equal-volume buckets, one pure white and one pure red. Pour a quarter bucket of red into white, mix, then pour the same volume of the mixture back into the red bucket and mix again. Which bucket's color changed more?

“I jumped straight to the independence formula and got a number, then the interviewer asked me to slow down and think about what the actual bounds were.”

View Post

Jane Street·Data Scientist·Technical Phone Screen

Jun 2026

Jane Street Data Scientist interview with a multi-part probability and game theory problem centered on a sealed-box pricing scenario. The question escalates from basic expected value through adverse selection and competitive bidding, so it's less a stats quiz and more a test of whether you can reason about incentives under uncertainty. Pretty intense for what I expected to be a standard quant screen.

A box contains a dollar amount drawn uniformly at random from $1 to $100. You can buy it at whatever price you name. What is the highest price you should rationally offer, and what is your expected profit if you buy at that price?
Now the dealer's profit equals the sale price minus the box contents. Is there any price where both parties benefit from trading? What does it mean if a dealer agrees to sell at a price you'd find attractive?
The company adds a $10 bonus to the dealer for every box sold. Now what price should you offer, and why?
The $10 bonus still applies, but now the dealer can open the box before deciding whether to accept your offer. You make a single take-it-or-leave-it offer. When does the dealer accept, what is your expected profit as a function of your offer price, and what offer maximizes your payoff?
Given that the dealer's ability to screen boxes destroys value compared to the no-peek scenario, what should the two parties agree to do instead?
Back to no one seeing inside the box, with the $10 bonus still in play. A second buyer shows up and competes with you for the same box. What happens to the price, and who captures the surplus?
Follow-up: how does your offer in the bonus scenario change if the dealer is risk-averse while you remain risk-neutral?
In the dealer-screens-boxes setting, if you interact with the same dealer repeatedly, can a 'always trade at a fixed price' agreement hold without an enforceable contract? What tempts the dealer to defect?
What happens if the two competing buyers in Part 4 can collude before bidding? How does the dealer's outcome change?

“Straightforward expected value warmup.”

View Post

Jane Street·Software Engineer·Technical Phone Screen

Jun 2026

Jane Street SWE interview with a pretty involved streaming data problem. The core question was well-designed and the follow-ups pushed hard on real engineering tradeoffs. No fluff, no behavioral stuff, just deep technical work the whole time.

You have a fixed set of M string codes and a stream of batches, each containing (timestamp, code, value) tuples in non-decreasing timestamp and lexicographic code order. Implement a stream transformer that emits one output row per timestamp, where each row is a length-M vector of values (or -1 if missing), ordered by the lexicographic code order.
Follow-up: within a given timestamp, the codes in a batch are no longer guaranteed to be in lexicographic order. How do you adapt?
Follow-up: a small fraction of records arrive with timestamps that are slightly out of global order. How do you handle this while keeping memory bounded and latency low?

“The base version is actually manageable once you see it clearly.”

View Post