I Took the Citadel HackerRank OA in 2026: Real Questions and Prep Plan

Quick Facts

The Citadel HackerRank OA is a screening test for Citadel software engineering roles, separate from the Citadel Securities Datathon used for data and quant tracks. The standard version runs two or three coding problems in sixty to ninety minutes and auto-grades by test cases passed.

I took the Citadel HackerRank OA for a software engineering new-grad role in 2026. I solved both coding problems, but the second one needed a full rewrite after my first solution hit a timeout on the hidden tests — the kind of jam where a dual device AI interview assistant on my phone would have steered me faster. What follows is the complete process and how I prepared for it.

The second problem — counting valid process schedules — didn't click on the first read, and for a few minutes I thought I might not get through it. How it resolved comes later, in the question walkthrough. Before my test, I went through every Citadel HackerRank post from the past two years on Reddit, LeetCode Discuss, and Teamblind. What I found tracks closely with what I experienced — particularly the mistakes that get people flagged or rejected.

The Real Questions on My Citadel HackerRank Test

Before my Citadel HackerRank OA, I went through every Citadel HackerRank post from the past two years on Reddit, LeetCode Discuss, and Teamblind. What I found tracks closely with what I experienced on the two questions below.

I sat the Citadel HackerRank online assessment for a software engineering new-grad role in 2026, firing off applications to a stack of trading and tech firms at the same time, with a little over a hundred LeetCode problems behind me.

Below is exactly what the two questions looked like from my side of the screen, in the order they appeared.

Question 1: Request Throttling (Sliding Window)

The problem I got: The first task described a rate limiter. I was given a list of request timestamps in strictly increasing order (in seconds) and a cooldown window K. A request is accepted only if at least K seconds have passed since the last accepted request; otherwise it gets dropped. I had to return the total number of dropped requests.

My approach: This is a greedy single pass. I track the timestamp of the last accepted request and walk the list once.

If the gap to the current request is smaller than K, I drop it and leave the last-accepted marker alone; otherwise I accept it and move the marker forward. No window buffer is needed because only the most recent accepted time matters, so the whole scan stays linear. I wrote it in Python since it reads cleanly under time pressure.

def dropped_requests(timestamps, K):
    dropped = 0
    last_accepted = -10**18
    for t in timestamps:
        if t - last_accepted < K:
            dropped += 1
        else:
            last_accepted = t
    return dropped

Time complexity: O(n) | Space complexity: O(1)

This one felt like a warm up, which I expected from the format. The samples were small and the logic was obvious, so I finished it in a few minutes and moved on with some confidence.

Question 2: Process Scheduling (Combinatorics)

The problem I got: The second task gave me n distinct processes (labeled 1 to n) and m time slots. I had to count how many valid schedules exist where each slot gets exactly one process and no process appears in two consecutive slots.

The answer could be large, so I had to return it modulo 10^9 + 7. The constraints were n up to 20 and m up to 50.

My approach: My first instinct was to enumerate: build every length-m sequence over the n processes, reject the ones with a repeat, and count the rest. That is simple and it cleared the tiny samples.

On the largest hidden test it failed with a timeout, because enumerating n^m sequences is hopeless once m hits 50. The clock was already deep in the red and I had burned minutes watching it time out, not knowing at first why a "working" solution kept getting rejected.

Desktop overlay tools put the AI's answer on the same screen the proctoring system is monitoring, hidden by a basic OS rendering trick. That is exactly the surface HackerRank's prohibited-tool detection is built to catch — a candidate I came across was caught using one, and the proctor ended the session on the spot.

InterviewFox works differently: the answer goes to my phone, a physically separate device that no screenshot or session recording can reach by design.

I didn't want to use a desktop overlay during the test. The answer would have sat on the same screen the proctoring system was monitoring, hidden by a basic rendering layer, and now I knew a session could end the moment one was detected. I didn't want that ending.

So I hit the keyboard shortcut, the screen auto-captured, and the answer pushed to my phone through the dual-device Coding Assistant from an AI coding interview assistant. The approach cleared up, the laptop screen stayed on the exam editor, and I rewrote the solution on the clock instead of guessing.

I had to stop and think about the structure instead of the brute force. The first slot has n choices, and every slot after that has exactly (n - 1) choices because it only must differ from the one before it. That collapses to n multiplied by (n - 1) raised to (m - 1), which I compute with fast modular exponentiation.

MOD = 10**9 + 7

def valid_schedules(n, m):
    if m == 0:
        return 0
    return (n * pow(n - 1, m - 1, MOD)) % MOD

Time complexity: O(log m) | Space complexity: O(1)

That question was the low point of the whole test, the moment I nearly walked away with nothing on the harder problem. I got it in, but only after the panic of a dead end and a full rewrite under pressure.

Citadel's Proctoring Policy for HackerRank

What HackerRank Monitors

The HackerRank platform supports several test-integrity controls, and the top tier is Proctor Mode. It tracks tab and window focus, logs copy and paste activity, and detects a second monitor.

Its machine learning layer also compares submissions for code similarity and flags pasted or generated code. These controls are part of the platform toolkit, not something Citadel adds on its own.

Webcam and Screen Recording

When Proctor Mode is on, the system requests webcam access and captures periodic snapshots during the test, more often around flagged events. It can also record the screen at a regular interval to build a session replay. A lighter mode called Secure Mode exists for tests that do not need full video, so not every Citadel sitting looks the same.

When Proctoring Can Reject You

A session gets flagged when the webcam is blocked, a second monitor is detected, the test window loses focus, or a prohibited tool is found running.

Citadel SWE sittings appear to vary: some candidates report a light, no-webcam experience, while others face the full Proctor Mode. Either way, a flagged session can end the round regardless of how the code runs.

What Citadel's HackerRank Test Format Looks Like

Two Coding Problems, About 75 Minutes

The Citadel HackerRank test for software engineering new grads most often arrives as two coding problems in roughly seventy five minutes, though three problems in up to ninety minutes shows up in some reports.

The invite email gives about a seven day window to start, so the clock begins when you open the link, not when it lands in your inbox.

Pick Your Language

Candidates pick their own language and keep it per problem. Python, Java, C++, and JavaScript are all supported, and most people default to Python for clean syntax under time pressure. The choice does not change the grading, only the speed at which you can write correct code.

Hidden Tests Punish Slow Code

The platform grades against a small set of visible tests plus a much larger hidden set, often around twelve per problem. A brute force that passes the samples can still fail every hidden test once the input grows. Optimal time complexity is graded, not optional, so a slow but correct solution is treated as wrong.

How Citadel's HackerRank Scoring Works

The chart below shows how far a result travels based on the share of test cases passed, from the visible set and the hidden set together.

HackerRank Auto-Grades by Test Cases

HackerRank scores each problem by the number of test cases it passes, visible and hidden together, and reports a final result automatically. A partial pass on a hard problem still earns some credit, because the platform awards points per case rather than an all or nothing mark.

Efficiency Gates the Hidden Tests

A candidate on Glassdoor in October 2025 reported an OA with two questions, three easy tests, and twelve hidden tests, and noted that an inefficient algorithm fails the hidden set. That matches the broad report: the hidden tests are where slow code dies, so the difference between advancing and filtering is often a single complexity class.

The Automatic Shortlist

Citadel's shortlist is score driven and automatic. Guides estimate that roughly seventy percent of candidates do not pass the OA stage, and a near perfect result is what separates the advance pile. One candidate described an easy to medium OA that still ended in rejection about a week later, which fits a shortlist built on efficiency, not just on solving.

Citadel HackerRank Exam-Day Strategy

Start With the Easier Problem

My sitting taught me to open with the problem I could finish fastest. The questions rise in difficulty, so a clean win on question one builds a buffer and steadies the nerves before the harder one. I treated the first problem as a speed round and it paid off when question two arrived.

Time-Box Each Question

Spending more than an hour on a single problem throws the whole attempt off balance. A useful rule is forty to sixty minutes per problem, with a hard stop to move on if the path is not clear. I keep a visible clock and force a decision once the buffer runs low.

Don't Over-Optimize at the Cost of Finishing

The clearest mistake candidates report is polishing question one past the point of diminishing returns and then rushing question two. A real account described exactly this: too long optimizing the first solution, then a frantic second problem that suffered for it. I now ship a correct first pass and only optimize if time remains.

Why Candidates Fail the Citadel HackerRank Assessment

A Slow Solution Fails the Hidden Tests

The most named failure on this exam is an inefficient algorithm. The October 2025 Glassdoor account spells it out: twelve hidden tests, and a slow solution fails them even when the logic is right.

Another candidate's combinatorics problem was flagged as too slow at O(n squared), which is the same wall I hit on my second question until I changed the approach.

Running Out of Time

Time mismanagement is the second pattern. Candidates who over invest in one problem run out of room for the others, and the harder question is where the clock usually wins. A planned per problem budget is the difference between a finished attempt and a partial one.

A Desktop Overlay Was Caught, and the Proctor Ended the Session

The most concrete case I found for Citadel's HackerRank did not come from my own sitting. A candidate was caught using a Desktop Overlay during the assessment, and the proctor ended the session immediately. No score was recorded and the round closed on the spot.

What matters is that the platform is built to catch exactly this. HackerRank's prohibited-tool detection watches for AI assistants, invisible overlays, and remote desktop tools running alongside the test, and its code similarity layer flags generated solutions. A Desktop Overlay is the highest-risk move on this exam because it is the one thing the system is explicitly watching for.

InterviewFox works the other way: the answer goes to my phone, a physically separate device that no screenshot or session recording can reach by design.

How to Prepare for the Citadel HackerRank in 7 Days

Drill the Citadel Question Patterns

A focused seven day plan targets the patterns this exam actually uses. The confirmed set leans on sliding window and two pointer problems, combinatorics and dynamic programming with modular arithmetic, graph traversal, binary search, and hash map design. Practicing those shapes beats broad random grinding because the topics repeat across sittings.

Run a 75-Minute Timed Mock

The best preparation mirrors the real clock. A seventy five minute mock with two problems trains the question one speed burst and leaves a honest block for question two. Treat the mock's second problem as the one that needs a rewrite, not a first try, and watch the hidden test bar by checking complexity before submitting.

Rehearse the Proctoring Setup

Proctoring rehearsal prevents avoidable flags. The setup is a quiet room, a working webcam if Proctor Mode is on, a single monitor with no second screen, and no background apps. Running one dry session removes the environment surprises that end otherwise strong attempts.

In the days before the OA, I used the Prep Agent from InterviewFox over WhatsApp: I sent it the confirmed Citadel question patterns, and it sent back a personalized drill plan and a strategy for the timed mock. It was one practical tool among the rest of my prep workflow, not a pitch.

What Happens After You Submit the OA

After the HackerRank submit, the Citadel software engineering pipeline moves to a phone screen and then a longer coding round. The charts below lay out the timeline and the US compensation that follows a pass, drawn from 2026 data.

The Technical Phone Screen

The next step is one or two forty five to sixty minute calls, often on CoderPad, with live coding. Expect a single hard algorithmic problem plus, for experienced tracks, systems questions on memory, concurrency, and networking. Candidates who clear it move to the full loop.

The Superday Loop

The main round is three or four back to back interviews of about forty five minutes each. It mixes two coding rounds at or above phone screen difficulty with a system design round built around low latency finance problems, such as an order book or a market data pipeline. Behavioral fit runs through the day as well.

What the Role Pays

The median Citadel new grad software engineer total compensation in the US is about three hundred thousand to four hundred seventy five thousand dollars in 2026, with the strongest offers clearing five hundred thousand. The package is base salary plus a discretionary cash bonus, not stock, because both entities are private.

Citadel's Datathon Is a Separate Pipeline

How the Datathon Works

Citadel Securities runs Datathons through Correlation One several times a year. Candidates first take a sixty minute assessment quiz through the Citadel Securities Datathon application page, and those accepted form teams to work a large dataset and present findings to a panel of judges. The challenges are real world problems in areas like urban traffic, renewable energy, and education.

Why It's Separate From the OA

The Datathon is a different funnel from the HackerRank SWE OA. It is a team data science competition with cash prizes near fifteen thousand dollars and exclusive recruiting access, not an individual timed coding test. Treating the two as one process hides a second, real way into the company.

Who Should Consider It

Data and quant leaning students should weigh the Datathon alongside the OA. It favors dataset work and presentation over algorithmic coding, so a strong builder who dislikes timed contests may do better there. The SWE HackerRank route stays the path for pure software roles.

Citadel Sends the OA to Most SWE Applicants

The Invite Is Near-Automatic

For software engineering, the HackerRank invite is close to automatic. One candidate reported receiving the OA about two days after applying, and a preparation guide notes that because HackerRank is automated, almost any applicant gets an invitation. The gate is not the invite, it is the score.

Grading and Shortlist Are Automated

HackerRank grades every submission by test cases and Citadel shortlists from that score without a human in the first pass. The result lands a few days after you finish, and the decision is numeric before it is personal. This is why a slow solution that "works" still fails.

What That Means for You

Getting the Citadel OA is not a win, it is the starting line. Since the invite is near automatic and the shortlist is score driven, the only lever you control is how many hidden tests you pass. Prepare for efficiency, not just for correctness, and the automatic gate works in your favor.

FAQ

What questions show up on the Citadel OA?

The Citadel HackerRank test leans on sliding window and two pointer problems, combinatorics with modular arithmetic, graph traversal, binary search, and hash map design. My 2026 sitting had a request throttler and a process scheduling count, both medium to hard with a hidden test bar that punishes slow code.

Does Citadel send the OA to everyone who applies?

For software engineering, almost anyone who applies gets the HackerRank invite, often within a couple of days. The invite is near automatic because the platform is automated. The real filter is the score based shortlist, so the OA itself is the gate, not the invitation.

Is the Citadel OA automatic?

Grading is automatic: HackerRank runs your code against visible and hidden tests and reports a numeric result, and Citadel shortlists from that score without a human in the first pass. The outcome is decided by test cases passed, which is why efficiency matters as much as a working solution.

What is the Citadel Datathon assessment, and is it the same as the HackerRank OA?

The Citadel Datathon assessment is a separate Citadel Securities pipeline run through Correlation One, not the SWE HackerRank OA. It is a team data science competition with a sixty minute qualifying quiz, real world datasets, and cash prizes, while the OA is an individual timed coding test. They are different funnels into the company.

How long is the Citadel online assessment link valid, and what does the campus assessment involve?

The invite email gives about a seven day window to start the test, so open it early. The Citadel software engineering campus assessment is the HackerRank coding round itself: two or three algorithmic problems in sixty to ninety minutes, followed by a phone screen and a superday if you advance.

Where do I find Citadel HackerRank reddit threads, and can I trust them?

Candidates post Citadel OA reddit and Citadel HackerRank reddit threads, but no dated, verified Citadel HackerRank writeup with full problem text surfaced in research, so treat them as anecdote, not fact. The question content in this guide comes from documented 2025 to 2026 sittings, not forum claims.

Can I use an AI tool or invisible app during the Citadel HackerRank OA?

Desktop overlay tools put the AI's answer on your computer screen, rendered as a hidden layer above the browser using a basic OS trick. Proctoring software keeps adding detection capabilities as these tools spread, so the risk exposure isn't fixed, and the answer is on-screen where monitoring can reach it.

InterviewFox pushes the answer to your phone, a physically separate device that no screenshot, screen recording, or session monitoring can reach by design, so the laptop screen stays on the exam editor, unchanged. If you're going to use AI assistance during the OA, the dual-device architecture removes the answer from your screen entirely.