Jane Street · Live Interview
Jane Street's first-round live interview is where strong written applications become offer pipelines or go nowhere. Below: the real questions Jane Street asks, what they're testing for, and how to practise live until it feels routine.
Practise for Jane Street
Start practising on Intervyo. Free tools, scored feedback, no payment.
The format
The live technical rounds bridge the early screen and the final-round superday, scheduled within days of a strong OA.
Two to three live rounds: a fast first screen, then deeper coding/math/market-making rounds, almost all 1-on-1.
Active practitioners: current quant traders, researchers, software engineers or desk leads.
Almost exclusively 1-on-1, conversational and interactive, not panel interviews.
Duration. First screen 30-45 minutes; subsequent rounds 45-60 minutes each.
Rounds at this stage. Typically two to three live technical rounds before the onsite.
Format breakdown
Phone, video, and in-person each have distinct mechanics. The interviewer scoring rubric is the same, but the operational preparation is different.
The 30-45 minute first screen (often phone or Zoom) is a high-velocity filter: under three minutes of resume, then straight into mental math (QT) or an algorithm (SWE), with zero margin for a slow or muddled baseline answer.
Deeper rounds run over video with a shared CoderPad or HackerRank Live editor (SWE/technical QT) or a shared doc/whiteboard for math derivations (QT/QR).
Advancing past these rounds leads to the in-person final round at the New York office, a continuous sequence of 4-5 back-to-back interviews.
Question categories
Question types cycled through the interview. For each, a real example, what the firm is screening for, plus weak and strong answer signals.
Bayesian updating, expected value and recursive/Markov states.
“You pull one of two coins (fair or double-headed) and flip heads. Probability it is the double-headed coin? After a second heads?”
What they test. Bayes and sequential updating
Weak answer. Probably more likely double-headed, so maybe 66%, then 75-80%.
Strong answer. P(D|H1) = (1 x 0.5) / ((1 x 0.5) + (0.5 x 0.5)) = 0.5/0.75 = 2/3. After a second heads, P(D|H1,H2) = (1 x 0.5) / ((1 x 0.5) + (0.25 x 0.5)) = 0.5/0.625 = 4/5 = 80%.
“Roll a fair die until a 6: expected number of rolls? Until two consecutive 6s?”
What they test. Geometric expectation and recursive states
Strong answer. For one 6, E = 6. For two consecutive 6s, set E0 = 1 + (1/6)E1 + (5/6)E0 and E1 = 1 + (5/6)E0; solving gives E0 = 42.
Protocol design and structured reduction of information-sparse puzzles.
“100 prisoners and one light switch: design a strategy to declare with certainty that everyone has visited.”
What they test. Designating a single coordinator to aggregate distributed events
Strong answer. Appoint one Counter and 99 Togglers. Each Toggler turns the switch on exactly once (the first time they find it off); the Counter turns it off and tallies. When the tally hits 99, everyone has visited.
Bypassing brute force with binomial expansions and shortcuts.
“Estimate 1.04 to the power of 10 to two decimal places in 15 seconds.”
What they test. Binomial/Taylor approximation under time pressure
Strong answer. Use (1 + x)^n approx 1 + nx + n(n-1)/2 x^2 with x = 0.04, n = 10: 1 + 0.40 + 45 x 0.0016 = 1.40 + 0.072 = 1.472, so about 1.47.
Risk arbitrage, time value and market-making mechanics.
“Company X is $100; a binding $130 cash offer closes in six months. Where does it trade, and why not $130?”
What they test. Risk arbitrage, time value of money and deal risk
Weak answer. It trades at $130 because that is the promised price.
Strong answer. It gaps up but trades at a discount, around $120-125, reflecting the time value of capital locked for six months and the non-zero probability the deal breaks (regulatory, financing, MAC), which would send it back toward $100. The price is the probability-weighted expected value.
Optimal data structures for latency-sensitive problems, stated complexity first.
“Design an order-matching engine supporting add_limit_order(side, price, quantity) and get_best_bid_and_ask().”
What they test. Optimal data-structure selection and edge cases
Weak answer. A single flat array sorted on every insert (O(N log N) per add).
Strong answer. Separate sides: a max-heap (or BST) for bids and a min-heap for asks keyed on price, each node pointing to a FIFO queue at that level. add_limit_order is O(log M) and get_best_bid_and_ask reads the roots in O(1).
Technical depth
The baseline is quantitative fluency, but the technical bar shifts sharply by track.
Mental-math speed, rapid estimation, probability/EV under pressure and live market-making games; how fast you update beliefs and manage risk.
Deep rigor: derivations from first principles, linear algebra (eigenvalues, SVD), regression, time-series and ML optimization.
LeetCode Medium-to-Hard algorithms plus system design, memory management, concurrency, race conditions and cache locality; OCaml not required but functional reasoning is valued.
Structured business reasoning and communication on open-ended commercial and market-structure problems, not high-speed math or coding.
The rubric
The interviewer is checking against a scorecard. Knowing the categories is half the battle.
Aggregation. You do not need every answer perfect: a candidate who makes a minor slip but shows speed, structure and swift coachability often outscores a slow, silent, correct one.
Pass threshold. A truly disastrous round (freezing, failing a baseline math check, or dogmatic dishonesty) almost always sinks the application; a single mixed round can be offset by exceptional other rounds.
Weighting vs other rounds. Technical competence is a prerequisite, but communication clarity and intellectual honesty carry immense weight; an arrogant or uncoachable candidate is discarded regardless of raw metrics.
How to practise
Conversational mocks that read your Resume, ask Jane Street-specific questions, push back when answers are weak, and produce a scored report.
Lazard · IB Analyst
Technical Interview · Resume-aware
Interviewer
I see from your Resume you completed Insight Program at Lazard. Walk me through a deal you observed during that placement.Interviewer
You mentioned trading comps. Walk me through how the peer set was selected, what criteria were used, and were there any borderline cases?Full report when you end
Why candidates fail
Specific failure patterns documented across past applicants. Most are avoidable with focused preparation.
Going completely silent
Falling silent for more than 15 seconds is catastrophic; the interviewer cannot grade reasoning they cannot hear.
Defending a flawed model too stubbornly
Arguing for an answer after an edge case or blind spot is exposed signals a lack of intellectual honesty.
Caving too fast without review
Panicking at "Are you sure?" and discarding sound work for a random guess.
Sloppy arithmetic under stress
Strong mathematicians failing on basic fraction or percentage errors under time pressure.
Treating market making as pure math
Holding bad quotes because "the fair value is 3.5" while ignoring inventory limits and adverse selection.
Freezing on ambiguous prompts
Complaining about missing data on a Fermi question instead of stating assumptions and building a framework.
What works
Concrete moves drilled by candidates who clear the cut, drawn from applicant accounts and recruiter feedback.
Continuous voice narration
Verbalize the mental trajectory constantly so the interviewer can follow even mid-calculation.
Explicit assumption stating
Take control of ambiguity: "I will assume a uniform distribution as a baseline, then adjust for skew."
Proactive sanity-checking
Test final values against boundary cases (N=0, N=infinity) out loud before moving on.
Graceful error correction
Stop, acknowledge calmly, and backtrack to the exact step rather than collapsing.
Strategic clock management
Switch to a first-order approximation to keep exploring the broader problem when bogged down.
Risk-first market making
Prioritize inventory management and spread discipline over chasing mathematical perfection.
From past applicants
Anonymised candidate accounts of how recent Jane Street applicants handled the live round. Each covers prep, the experience, and the outcome.
Quantitative Trading (Princeton undergraduate)
Prep. Drilled mental math and conditional probability to automaticity.
Experience. Two minutes of resume, then three consecutive mental-math questions and a conditional dice problem; made an arithmetic slip but narrated the correction immediately, then ran a live market-making game on US Olympic gold medals, getting forced into a heavy short and widening the spread while narrating inventory risk.
Outcome. A superday invite less than 24 hours later.
Software Engineering (UC Berkeley undergraduate)
Prep. Comfortable stating complexity before coding and reasoning about hardware.
Experience. A CoderPad graph-traversal problem (routing an order across dark pools); stated an O(V^2) brute force, explained why latency made it unacceptable, then optimized to O(E + V log V) with a Dijkstra variant, before pivoting to lock-free atomics and cache-line alignment under a concurrency curveball.
Outcome. Advanced to the final round.
What gets you through
FAQ
Live coding interviews primarily use CoderPad or HackerRank Live. You pick your language from a dropdown, and in the vast majority of software engineering and technical quantitative rounds you are expected to compile and run your code against active test cases during the interview to prove correctness. State your time and space complexity in Big-O before writing a line, write clean modular functions with meaningful names, and handle edge cases like empty inputs, null pointers and integer overflows explicitly. The interviewer will not penalize Python over C++, but they will judge whether you use your chosen language idiomatically.
Do not guess blindly and do not pretend. Acknowledge the boundary immediately and pivot to first principles: "I have not studied this specific distribution, but based on standard random walks I will set up a recursive state equation." Jane Street scores intellectual honesty and self-correction very highly, so if you spot a flaw in your own math or code mid-answer, interrupt yourself, acknowledge it calmly and fix it. That composure under pressure reads far better than defending an invalid model or freezing in silence.
You must quote a valid two-sided market (bid strictly below ask) on an unknown or evolving quantity, such as the sum of hidden dice or the number of windows in a building. The interviewer buys from your ask or sells to your bid, often holding asymmetric information, and feeds you new data as the game runs. The skills tested are inventory management (lean your whole market down when long, up when short), adverse-selection awareness (if someone keeps trading with you, your price is likely wrong, so shift the mid-market toward the flow), and spread discipline (wider when uncertain, tighter as you gain information) while tracking your running P&L. Intervyo can simulate these games with live, adaptive counterparties and instant feedback.
The other rounds
Live interview is one of four rounds. Practise each one free on Intervyo.
Practise free
Practise every stage on Intervyo with AI-scored feedback: HireVue, psychometrics, live mock interviews, and Resume. Free to start, no card required.
Start practising freeFree tools, upgrade any time
Intervyo is not affiliated with or endorsed by Jane Street. Interview questions are sourced from past applicants and the firm's published guidance; verify on the firm's careers site. Sector: Investment Banking.
Jane Street
Practise free