Algorithmic coding (Cubist QR, Quant Dev, SWE)
3 to 4 problems · 90 to 120 minutes
What it tests. Converting math and logic constraints into optimized code under execution bounds.
Worked example. Greedy array problems, number theory and DSA wrapped in business or statistical contexts. One worked example: given an array, you may swap each adjacent pair at most once to maximize the position-weighted sum; the net gain from swapping arr[i] and arr[i+1] reduces to arr[i] minus arr[i+1], so swap greedily whenever the left element is larger. For [2,1,4,3] this yields a maximum of 30.
Common traps. Over-engineering with complex patterns; neglecting empty arrays, integer overflow or negative bounds; ignoring the constraint sizes.
How to handle it. Read constraint boundaries first. If N can reach 10^5, an O(N^2) solution triggers a time-limit-exceeded error; you need O(N log N) or O(N). Outline logic in comments before coding.
Mental math, arithmetic speed and number sequences
Strict per-item timers
What it tests. Working-memory stability and high-speed processing under stress.
Worked example. Convert fractions to decimals instantly: seven-eighths of 160 is 140, and a quarter of 44 is 11, so the expression equals 129. For sequences, compute first and second differences and check for alternating sub-sequences.
Common traps. Scratchpad calculations on items that need mental shortcuts; one error cascading into missed questions on a rigid timer.
How to handle it. Use complementary numbers and rounding. Drill fraction-to-decimal conversions and complementary subtraction daily.
Probability, statistics and combinatorics (Cubist)
45 to 60 minutes
What it tests. Conditional probability, distributions, expectation operators and discrete structures at upper-undergraduate to graduate level.
Worked example. A Bayes problem: a signal with 95% sensitivity, a 2% false-positive rate and a 0.5% base rate of a crash gives a posterior of roughly 19.3% that a crash is occurring when the signal fires. A combinatorics problem: distributing 10 identical risk units across 4 strategies, each at least 1 and strategy 1 capped at 3, gives 64 valid allocations via stars-and-bars with inclusion-exclusion.
Common traps. Confusing unconditional and conditional expectations; breaking independence in multi-stage sampling.
How to handle it. For sequential updates, write out Bayes or a joint-probability matrix rather than trusting intuition.
Finance, accounting and valuation (Academy)
45 to 60 minutes
What it tests. Three-statement mechanics, working-capital drivers and core valuation metrics.
Worked example. A $100 software asset amortized straight-line over 5 years at a 20% tax rate: net income falls $16, operating cash flow rises $4 (adding back $20 of non-cash amortization), and the balance sheet nets to a $16 decrease on both sides, staying balanced.
Common traps. Forgetting that working-capital changes have inverse cash effects; mis-adjusting net income for non-cash items.
How to handle it. Hold the identity assets equal liabilities plus equity, and trace every income-statement change through cash flow to the final cash line.
Markets and the written investing case (Academy)
48 to 72 hours from prompt activation
What it tests. Commercial evaluation and thesis structuring on a mid-cap ($2B-$10B) you understand.
Worked example. A short thesis on a hypothetical mid-cap: consensus models 15% revenue growth at 18x forward EV/EBITDA; the variant view is that customer-acquisition cost has risen 34%, pushing the LTV:CAC ratio below 1.8x and flattening growth to 2%. The catalyst is the next earnings release within 90 days; the asymmetric risk / reward is roughly 3:1.
Common traps. Choosing an efficient mega-cap where edge is near-impossible; a long thesis on past performance with no forward catalyst.
How to handle it. Build from unit economics, not high-level growth averages, and state a concrete 6-12 month catalyst timeline.
Personality / behavioral questionnaire
What it tests. Risk preferences, behavioral consistency under loss and cognitive openness; a secondary indicator, not a hard filter.
Worked example. Forced-choice statements: the most-descriptive choice should signal a systematic, data-driven approach under uncertainty; the least-descriptive should be acting on uncalibrated intuition over structured analysis.
Common traps. Selecting contradictory extremes to look optimized, which triggers low-consistency integrity flags.
How to handle it. Answer authentically as a calibrated, accountable risk-taker who updates on conflicting evidence and works within strict boundaries.