Headlands is a systematic proprietary trading firm founded by former senior Citadel traders, headquartered in Chicago with additional offices including Amsterdam, London, and New York. It trades its own capital in equities, options, and futures rather than managing outside money, which shapes the interview: there is no client-facing polish to screen for, only whether you can find edge and whether the code that captures it runs fast enough to matter. One point of confusion worth clearing up before you research comp: TD acquired a separate Headlands fixed-income unit in 2021, but the proprietary trading parent remains independent.
That technology-first posture is the thing to internalize. At a lot of shops the researchers invent signals and hand them to engineers. Headlands treats the low-latency stack as first-class, so a market-data handler that shaves microseconds is valued alongside a good signal. Even on the research track you will get asked about spreads and adverse selection, and even on the developer track you will get asked about probability. The loop below reflects that overlap.
The online assessment: C++ coding plus statistics
The first real gate is a timed online assessment, typically on HackerRank. Candidates consistently report two components: an algorithmic coding problem you solve in C++, and a set of statistics and probability questions. The coding half is reported as the easier part, roughly a medium-to-hard LeetCode-style problem in a single sitting (candidate reports describe a window of about 90 to 100 minutes). If your data-structures fundamentals are current, this is not where people wash out.
The probability and statistics half is where prep actually pays off. Expect expected-value setups, conditional probability, and distributions, and expect several of them to wear a markets costume rather than a textbook one. Practical examples in the style people report:
- A stock price takes independent one-tick steps, up or down with equal probability, each second. After ten ticks, what is the probability it sits at least four ticks above where it started, and what is the expected size of the absolute move?
- A one-day future settles at the number of heads in ten fair flips. What is its fair value, and where would you set a bid and an offer if you had to make a two-sided market on it?
- You flip a fair coin until you see two heads in a row. What is the expected number of flips?
None of these are exotic, but they punish rustiness. If your linearity-of-expectation and indicator-variable reflexes are sharp, you move through them quickly; if you are deriving from scratch under a clock, you run out of time. The random-walk and fair-value framings matter because that is how the desk actually thinks about price.
Technical phone screen
Passing the assessment usually leads to one or two technical phone screens, 45 to 60 minutes each. For every track this includes live coding, and the interviewer is watching how you reason out loud, not just whether the tests pass. Developer-track screens push harder on language specifics and complexity trade-offs; research-track screens weave in a probability or market-intuition question between coding segments. Because the firm runs lean, you tend to talk to fewer people than at a large bank, and each conversation carries more weight.
The final round, split by track
The final round is where the QR/QT and developer paths visibly diverge. It is typically three to five sessions, virtual or onsite, run by the people you would actually work with.
Quant researcher and trader focus
Researchers and traders spend most of the loop on probability, expected value, and the reasoning behind market making. The brainteaser-style questions are a proxy for how you handle uncertainty and structure a messy problem. Representative phrasings:
- You roll a fair die repeatedly and add the results. What is the expected number of rolls until the running sum first exceeds 12?
- You quote a two-sided market and your order flow is 60% informed. How do you adjust your spread, and what happens to your inventory if you get it wrong?
- You are offered a bet at fixed odds on a repeated game. What edge do you need before you size up, and how does variance change the answer?
Those last two are market-making reasoning in disguise: spread, inventory risk, and adverse selection. Strong candidates connect the math to the trading consequence rather than treating each question as an isolated puzzle. For experienced QR and QT hires, expect at least one session to shift from puzzles to your own track record, a “walk me through a strategy you built and how you validated it” conversation that probes how you guarded against overfitting and decided the signal was real.
Software and quant-developer focus
The developer track keeps the algorithmic coding at a solid bar and then goes deep on C++ and systems. This is where the low-latency culture is most obvious. Expect language questions that only matter if you have written performance-sensitive code:
- Contrast the memory layout of
std::vectorandstd::deque, and explain when cache locality makes the vector faster despite an O(n) insert. - Where does a
std::shared_ptrcontrol block cost you on a hot path, and how would you avoid atomic refcount traffic there? - Implement a single-producer, single-consumer lock-free ring buffer, and explain the memory ordering you need.
Systems design prompts come straight off the trading path rather than from generic web-scale templates. A common shape: design a market-data handler that parses exchange feeds and maintains an order book with the lowest possible latency, or design an order-management path that stays correct under bursty load. They want to hear about cache-aware data-structure design and what you would measure, not a box diagram. Keeping your Big-O and complexity fundamentals automatic frees you to spend the conversation on the parts that are specific to trading systems, and working through a few system design interview guides beforehand helps even though the prompts here are narrower than the web-scale classics. Senior developer loops add lock-free concurrency and more pointed questions about contention and false sharing.
Round structure at a glance
| Stage | Format | QR / trader focus | Software / quant-dev focus |
|---|---|---|---|
| Online assessment | Timed, HackerRank, ~90-100 min | Statistics and probability plus one C++ coding problem | C++ coding problem plus statistics questions |
| Technical phone screen | 1-2 calls, 45-60 min each | Live coding with a probability or market question | Live coding with language and complexity depth |
| Final round | 3-5 sessions, virtual or onsite | EV, brainteasers, market-making, strategy evaluation | Algorithms, deep C++, trading-systems design |
| Decision | Fast turnaround | Offer commonly within 1-2 weeks of the final round | |
What each round is actually screening for
The assessment filters for baseline coding plus live probability recall, so it front-loads the two skills that are cheapest to test at scale. The phone screen checks that you reason cleanly under mild pressure and that your online performance was your own. The final round is where judgment gets tested: can a researcher tie a probability result to a trading decision, and can a developer make correct latency trade-offs and defend them. Because the team is small, cultural and collaboration signal is read continuously rather than in a separate behavioral loop.
Compensation: how to think about the numbers
Headlands is a top-tier systematic prop shop, and pay tracks that tier rather than a standard tech-company band. Compensation is base salary plus a discretionary performance bonus, and at prop firms the bonus is the part that moves, sometimes a large multiple of base for strong years, because the firm is sharing trading P&L rather than paying against a fixed budget. Public reporting for new-graduate quant and developer roles at top-tier systematic prop shops commonly lands in a broad range around $250,000 to $500,000 in first-year total compensation, with base salary often from the low six figures to roughly $200,000 and the remainder in bonus and sign-on. Treat that as a wide band, not a quote: the bonus dominates and swings hard with desk and firm performance, and experienced traders and researchers can clear it by a lot in a good year.
Rather than anchor on any single number, treat base as the stable floor and the bonus as the variable that depends on desk and firm performance, then cross-check current, crowd-sourced data points on levels.fyi and model the split with a total-comp calculator before you compare offers.
How to prepare
For the research and trading track, the highest-leverage prep is rebuilding fluency in probability and expected value until indicator variables, linearity of expectation, and conditional-probability setups are reflexive. Practice speaking the market-making logic out loud: what a spread compensates you for, why inventory is a risk, how adverse selection eats an edge. Studying how other quant firm interview guides phrase probability and EV questions is a fast way to build that pattern recognition, since the same setups recur across desks. If you can only sharpen one thing, make it timed expected-value problems in a markets framing.
For the developer track, keep algorithmic coding warm but spend the marginal hour on C++ that you rarely touch in application code: manual memory management, STL container internals and their cache behavior, and lock-free patterns if you are targeting senior levels. Be ready to design a piece of the trading path and to talk about what you would profile. The coding patterns cheat sheet covers the algorithmic baseline the assessment expects.
Both tracks benefit from studying adjacent loops, because the systematic-prop playbook rhymes across firms. The Hudson River Trading, XTX Markets, D. E. Shaw, and Millennium guides show the same probability-plus-low-latency pattern from different angles. The through-line at Headlands specifically: they will not let you separate the math from the machine, so prepare to defend both in the same breath.
