Citadel · Live Interview
Citadel's first-round live interview is where strong written applications become offer pipelines or go nowhere. Below: the real questions Citadel asks, what they're testing for, and how to practise live until it feels routine.
Practise for Citadel
Start practising on Intervyo. Free tools, scored feedback, no payment.
The format
The live technical gauntlet sits between the early screen (OA plus recruiter call) and the final-round superday.
2-4 distinct rounds, mostly 1-on-1 with active practitioners, with occasional 2-on-1 panels; CoderPad or live HackerRank for coding, interactive whiteboards for math.
Current QRs, QTs, senior SWEs and desk heads - not HR.
Interviewers probe weak answers. The candidates who get through handle follow-ups confidently, not just the headline question.
Duration. 45-60 minutes each (a first video screen may be 30-45).
Rounds at this stage. 2-4 rounds, often scheduled back-to-back within a single week.
Format breakdown
Phone, video, and in-person each have distinct mechanics. The interviewer scoring rubric is the same, but the operational preparation is different.
A 30-45 minute video first round filtering people who look strong on paper but cannot execute; QT leads with 5 minutes of rapid mental math or probability.
Conducted on Zoom or Teams with shared CoderPad or live HackerRank editors; QR/SWE move straight into coding or a shared math canvas.
Senior laterals and some tracks come onsite to Miami, New York or Chicago for consecutive 45-60 minute physical-whiteboard rounds.
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.
First-principles derivations shown step by step, not jumps to a final form.
“You start at 0 and stop at +3 or -1 on a fair coin walk. Probability of stopping at +3, and expected flips?”
What they test. Random walks, absorption probabilities, martingales
Strong answer. Linear absorption gives P(reach +3 from 0) = (0+1)/(3+1) = 1/4 = 25%; solving the expected-steps recursion E(x) = -x^2 + 2x + 3 gives E(0) = 3 flips.
“Expected value of the maximum of three independent U(0,1) variables?”
What they test. Order statistics and integration
Strong answer. The CDF of the max is y^3, the PDF is 3y^2, and the expected value integrates to 3/4; in general the expected max of n uniforms is n/(n+1).
“Polya urn: 4 red, 6 blue; draw, replace with 2 more of that color, draw again. Given the second is red, probability the first was red?”
What they test. Bayes theorem and conditional expansion
Strong answer. P(R2)=2/5 and P(R2|R1)P(R1)=1/5, so P(R1|R2)=1/2; the marginal probability of red on any later draw stays at the initial 2/5 proportion.
Problems with no textbook formula; reduce to a base case and find the pattern.
“The 100 prisoners and 100 boxes problem - best strategy and survival probability?”
What they test. Permutation cycle decomposition
Strong answer. Follow your own loop: open the box with your number and chase the cycle. You all survive unless a cycle exceeds 50, so survival is about 1 - ln(2) = roughly 31%, vastly above (1/2)^100.
“Insert 5 letters into 5 addressed envelopes at random - probability none is in its correct envelope?”
What they test. Derangements and inclusion-exclusion
Strong answer. The number of derangements !5 = 44, so the probability is 44/120 = 11/30 = roughly 36.67%, converging to 1/e as n grows.
Speed under stress, narrated out loud.
“What is 43 times 47? Five seconds.”
What they test. Algebraic pattern recognition and speed
Strong answer. They are symmetric around 45: (45-2)(45+2) = 2025 - 4 = 2021.
“Estimate the weekly dollar volume of 500mL bottled water sold in NYC.”
What they test. Structured Fermi estimation
Strong answer. Population (~8M) times a weighted ~2.25 bottles per person times ~$1.50 gives roughly $27M per week, sanity-checked against a ~$1.4B annual market.
Genuine commercial reasoning, not memorized headlines.
“A $100/share cash acquisition closes in six months; the stock trades at $94.50. Why the discount, and is it a good trade?”
What they test. Merger arbitrage, time value and deal risk
Strong answer. The $5.50 spread reflects deal-completion risk and time value: a ~5.82% six-month return, ~11.6% annualized, judged against the risk-free rate plus a premium for the probability of a regulatory block.
“Stock at 100, one-month 105 call at 2, rates 0, no dividends - price the 105 put.”
What they test. Put-call parity
Strong answer. With r=0, C - P = S - K, so 2 - P = 100 - 105 = -5, giving P = 7; any other price opens a risk-free arbitrage.
Clean, compiling, optimal code with stated complexity before the first line.
“Maximum subarray - largest contiguous sum, with complexities stated.”
What they test. Kadane algorithm, edge cases, clean syntax
Strong answer. Initialize current and global max to the first element to handle all-negative arrays, then run Kadane in O(n) time and O(1) space.
“Design an electronic limit order book supporting insert, cancel and market-order execution efficiently.”
What they test. Data-structure selection for O(1) or O(log n) operations
Strong answer. Discuss the right structures per operation and state the complexity each achieves before coding.
The signature QT exercise: price uncertainty, manage inventory, decode the counterparty.
“I rolled two dice you cannot see; make a continuous market with max spread $1 and a +/-3 position limit.”
What they test. Adverse selection, inventory skew and information updating
Strong answer. Open centered on EV 7.0 (6.5 at 7.5); after repeated buys, skew the whole market up and lift your EV estimate; at the -3 limit, quote only to be hit on the bid to flatten, and read a pass as information about the true value.
“Design a 40% guaranteed annualized return with zero risk, zero capital, no leverage or derivatives.”
What they test. Composure on an impossible prompt and intellectual honesty
Strong answer. Calmly state that no-arbitrage makes this impossible in efficient liquid markets; anyone claiming otherwise is hiding embedded tail risk.
Technical depth
Every track faces a baseline of math rigor and algorithmic competence, but each is pushed to an extreme bar in its domain.
Derive estimators and prove properties on a whiteboard (Taylor expansions, Ito-based SDEs, bias and variance of non-linear estimators), formulate loss functions explicitly (Ridge from matrix calculus, L1 vs L2 as Laplace vs Gaussian priors), and master stationarity, cointegration, ARMA/ARIMA and GARCH.
Process multi-step arithmetic while talking, size bets with the Kelly criterion, and show instinct to widen spreads or cut size when variance rises or you suspect you are being picked off.
Flawless LeetCode Medium/Hard with optimal complexity, CPU cache hierarchies and false sharing, branch prediction, memory-mapped files, and lock-free concurrency (for example a thread-safe SPSC ring buffer with explicit memory ordering).
Defend a long/short stock pitch against a skeptical PM, walk working-capital changes through the three statements, build a multi-scenario DCF and weigh ROIC against WACC.
The rubric
The interviewer is checking against a scorecard. Knowing the categories is half the battle.
Aggregation. Interviewers log narrative feedback and scores, then hold a synchronization meeting to decide on the superday.
Pass threshold. Roughly 20-30% advance per round; a clear fail in a core technical category generally ends the pipeline, though historic brilliance can earn a tie-breaker round.
Weighting vs other rounds. Technical skill is a non-negotiable prerequisite - eloquence cannot save wrong math - but between two equal candidates the clearer communicator who handles feedback gracefully wins every time.
How to practise
Conversational mocks that read your Resume, ask Citadel-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.
Black-box silence
Staring at the screen working everything in your head for minutes leaves the interviewer no window into your reasoning.
Defending a wrong answer too hard
Arguing against a clear counter-example signals ego over data and is an immediate disqualifier for intellectual honesty.
Caving instantly under a nudge
Abandoning a correct proof the moment you are asked "are you sure?" shows a dangerous lack of conviction.
Treating market-making games as pure math
Perfect static EV but no inventory adjustment when repeatedly hit leads to an unhedged blow-up.
What works
Concrete moves drilled by candidates who clear the cut, drawn from applicant accounts and recruiter feedback.
Continuous verbal redirection
Narrate as you compute: "I am setting up a three-state Markov chain, but first let me check the transition probabilities sum to 1."
Explicit assumption mapping
State assumptions up front (discrete-time symmetric walk) and offer alternatives if the interviewer prefers.
Proactive sanity-checking
Run extreme cases (n=0, p=1) to confirm a formula collapses to a known bound.
Graceful trajectory correction
When a flaw is pointed out, acknowledge it instantly and adjust the equations without defensiveness.
From past applicants
Anonymised candidate accounts of how recent Citadel applicants handled the live round. Each covers prep, the experience, and the outcome.
Quantitative Trader (Princeton graduate, Citadel Securities Miami)
Prep. Technical round scheduled two days after clearing the mental-math OA; conducted via Zoom with a senior SPX option trader.
Experience. Five minutes of rapid-fire fractional multiplication, squaring numbers like 76 and conditional dice probabilities. I missed one and simply said "correction, it is 2400 not 2600" and kept moving, then played a 35-minute market-making game on the number of aces in a modified deck while he fed in new information and traded against me.
Outcome. I never went silent, narrating my EV framework throughout, and advanced to the superday within 24 hours.
Quantitative Researcher (MIT PhD candidate, hedge fund)
Prep. Three separate 1-hour rounds on math and statistical inference, no LeetCode.
Experience. Derived a maximum-likelihood estimator for a custom distribution modeling regime shifts in LaTeX inside CoderPad, was pushed on bias under non-Gaussian noise, took an integration-by-parts hint cleanly, then derived the PCA projection matrix from scratch and proved its variance-maximization property.
Outcome. They wanted absolute mathematical rigor with no hand-waving; I advanced.
Software Engineer (CMU CS undergraduate, Citadel Securities)
Prep. Two CoderPad rounds with the compiler enabled for the core infrastructure track.
Experience. A LeetCode Hard graph problem with dynamic routing; I talked through a modified Dijkstra with a custom priority queue at O(E + V log V), wrote ~20 lines of clean C++, then got grilled on cache locality and concurrent edge-weight updates, explaining lock-free atomic pointers and cache-line bouncing.
Outcome. They cared as much about how the code interacts with the kernel and hardware as whether it works; I moved forward.
What gets you through
FAQ
Very fast. If your OA score sits in the top tier a recruiter usually contacts you within 48 hours, technical rounds are often scheduled back-to-back within a single week, and the whole path from OA to a final offer can conclude in under 14 days. Coding rounds run on CoderPad or live HackerRank sessions with real-time syntax highlighting and live compilation across C++, Python and Java, while probability and math rounds use a shared whiteboard or a LaTeX text editor where you show every intermediate step.
Do not guess wildly or pretend. State what you do know and start from first principles out loud: for example, "I have not seen this exact structure, but it looks like a geometric sequence - let me write the first three terms and look for a pattern." That demonstrates coachability and logical stability, which the rubric rewards. If you make an error and the interviewer flags it, acknowledge it immediately and correct course; both stubbornly defending a wrong answer and instantly caving on a correct one are scored as failure modes.
For Quantitative Research, Python is well accepted for clean vector math and modeling; for core-infrastructure SWE, clean C++ is favored so you can show low-level memory and systems knowledge. You do not need a flawless score in every round - a candidate who is exceptional in probability and market-making can advance even with merely average coding speed - but a catastrophic failure in a core domain (a SWE failing concurrency, or a QR with weak foundational calculus) generally ends the pipeline at the desk synchronization meeting. Intervyo runs realistic, firm-specific mock rounds with conversational follow-ups and instant feedback on accuracy, reasoning and delivery.
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 Citadel. Interview questions are sourced from past applicants and the firm's published guidance; verify on the firm's careers site. Sector: Investment Banking.
Citadel
Practise free