Optiver-Style Trading Assessment

How it works

The sitting runs six sections back to back, with the clock never stopping between them - about 85 minutes in total. It opens with 80 in 8, the headline round: 80 mixed arithmetic items (round-number shortcuts, percentage swaps, clean divisions, fraction-of, memorized squares and roots, two-step chains) in 8 minutes, roughly six seconds a question, marked +1 correct / -1 wrong / 0 skipped. Next is NumberLogic: 24 numeric sequences in 25 minutes, where you identify the rule (rising differences that cross zero, alternating multiply-and-adjust, two interleaved subsequences, or a second-order recurrence) and supply the next term, also marked -1 for a wrong answer.

Third is Likelihood List, new to the 2026 battery: 10 items in 8 minutes where you rank three outcomes - a coin-flip run, a dice total, a card draw - from most to least likely. This section carries no penalty. Fourth is Beat the Odds: 20 probability and expected-value items in 14 minutes covering die EV, ticket EV, all-dice-match probability, odds-to-probability conversion, and at-least-one complements, marked -1 for a wrong answer like the first two sections.

The final two sections carry no penalty. Zap-N runs 16 numerical working-memory items in 12 minutes - sum a block of five to seven numbers, or find the largest and multiply it by a given factor. Zap-Q closes the sitting with 18 abstract pattern and spatial items in 14 minutes: continue a repeating symbol cycle, count occurrences of a target shape in a grid, or spot which position breaks an otherwise uniform row.

No section allows going back once an item is left behind - you answer or skip, and the next item loads immediately. Every section shares that rule even though only three of the six penalize a wrong answer, so the real decision on the no-penalty sections is speed, not caution, while the real decision on the penalized sections is whether to answer at all.

How scoring works

Each section's raw score is correct answers minus wrong answers times that section's penalty; an unanswered item always scores zero, whether or not the section penalizes wrong answers. 80 in 8, NumberLogic, and Beat the Odds use penalty 1 (marked "+1 / -1 / 0" in the interface); Likelihood List, Zap-N, and Zap-Q use penalty 0 (marked "+1 / 0").

The sitting's overall score is the sum of every section's raw score (scaled), out of the sum of every section's item count (maxScaled = 80 + 24 + 10 + 20 + 16 + 18 = 168). The review screen reports this as a percentage: round(scaled / maxScaled x 100), floored at zero if penalties push the raw total negative.

Every section has allowBack set to false, so once you leave an item you cannot return to it within the run. You can flag an item for review with the F key or the flag button while answering it; since no section permits jumping back, a flag can only ever mean "show me this one on the marked script" - it surfaces on the final review screen, after every section is finished, with the prompt, what you entered, the correct answer, and the explanation.

80 in 8

At six seconds a question there is no time for long multiplication or column arithmetic - the items are deliberately built so a shortcut beats brute force. Watch for pairs that straddle a round number (a difference-of-squares setup like 64 x 76), percentages that are easier to compute swapped (15% of 240 equals 240% of 15, i.e. 15 x 2.4), divisions by a decimal that clear cleanly once scaled to whole numbers, fractions of a number chosen so the denominator divides it exactly, and memorized squares or powers of two.

Because this section penalizes a wrong answer, a guess is only worth it if you can eliminate enough of the arithmetic to be confident, not merely fast. If the shortcut has not clicked within a couple of seconds, skip - a skip costs nothing, a wrong guess costs a full mark on top of the one you didn't get.

NumberLogic

Every sequence here is generated from one of four rule families: differences that themselves increase (forcing the sequence to cross zero), an alternating multiply-then-adjust-by-one step, two interleaved subsequences moving at different rates, or a second-order recurrence where each term depends on the two before it. Write out first differences before anything else - if they are not constant, check whether they are themselves rising or falling in a pattern, and if nothing fits, split the sequence into odd and even positions and re-check each half on its own.

This section also penalizes a wrong answer, and 25 minutes for 24 items is generous compared to 80 in 8 - use the slack to verify your rule against every given term, not just the first two, before committing to the next one.

Likelihood List

There is no penalty here, but ranking by gut feeling is a trap by design: all six possible orderings of three outcomes are used as answer choices, so an intuitive-but-wrong ranking is always available to pick. Convert every outcome to an actual number before ranking - benchmark probabilities like a single die face (1/6), a coin-flip run (0.5 to the power of the flip count), a card suit or rank (out of 52), and a dice-sum count (out of 36) are all fast to compute exactly rather than estimate.

Since there is no downside to answering, never leave one of these blank - even a partial elimination (ruling out the outcome you are certain is smallest or largest) narrows six choices to two.

Beat the Odds

This section rewards a small set of formulas run correctly under time pressure: a fair n-sided die's expected value is (n+1)/2; a ticket's expected profit is its win probability times the payout minus its cost; the probability every one of k dice shows a chosen face is (1/6) to the power of k; betting odds of a-to-b against convert to an implied probability of b/(a+b); and the probability of at least one six in n rolls is 1 minus (5/6) to the power of n, not n/6 - a common error that double-counts.

Because a wrong answer costs a full mark here too, the fast checks are worth doing even under time pressure: an expected value can't exceed the largest possible payout, and a probability can't exceed 1 or fall below 0. Either check catches an arithmetic slip in under a second.

Zap-N

With no penalty and a working-memory format, the goal is throughput, not caution. For the summing variant, keep a single running total as you read each number rather than trying to hold all five to seven values in your head and add them at the end - one accumulator is far more reliable under time pressure than mental storage of a growing list. For the largest-value variant, scan once to find the maximum, then do the multiplication - don't recompute the scan after you've already spotted the largest number.

Answer every item. There is no cost to a wrong answer, and a rough estimate submitted with a second left beats a blank.

Zap-Q

Three distinct item types recur: a repeating symbol cycle (identify the period, usually three, and continue it), a grid-counting task (count occurrences of one target shape among distractors), and an odd-position task (every cell matches except one, and you name its position). None require more than direct observation - resist the urge to look for a subtler rule than the one actually generating the row.

This section also carries no penalty, so treat every item as worth attempting even under the clock; a fast, close guess on a counting item is better than a skip.

Pacing across the whole sitting

The clock never stops between sections, so a slow finish on 80 in 8 eats directly into NumberLogic's budget rather than resetting. Treat the six sections as three penalized rounds (80 in 8, NumberLogic, Beat the Odds) where skipping is a real, correct decision when unsure, and three free rounds (Likelihood List, Zap-N, Zap-Q) where answering everything is strictly better than leaving anything blank.

Because no section allows going back, use the flag key liberally on anything you are uncertain about rather than dwelling on it mid-section - flagging costs nothing, and the review screen surfaces every flagged item with the correct answer and explanation once the full sitting ends, which is the only chance you get to actually learn from it.

A worked example

1. Setup: 64 x 76 looks like an awkward two-digit multiplication, but both numbers sit an equal distance from a round number - recognize the straddle before doing any arithmetic.

   $$64 = 70 - 6, \qquad 76 = 70 + 6$$

2. Rewrite the product using the difference-of-squares identity, which trades a two-digit multiplication for two squares and a subtraction.

   $$64 \times 76 = (70-6)(70+6) = 70^2 - 6^2$$

3. Both squares are fast to compute from memory or a quick calculation.

   $$70^2 = 4900, \qquad 6^2 = 36$$

4. Subtract to get the final answer. Enter 4864.

   $$70^2 - 6^2 = 4900 - 36 = 4864$$

5. This section penalizes a wrong answer, so the shortcut only pays off if you trust it - if the straddle isn't obvious within a couple of seconds, skip rather than fall back to long multiplication against a six-second clock.

Common mistakes

• Grinding long multiplication on 80 in 8 instead of spotting the round-number straddle, percentage swap, or clean-division setup the item was built around - this alone burns the entire six-second budget on a single question.

• Guessing blind on 80 in 8, NumberLogic, or Beat the Odds when genuinely unsure. All three penalize a wrong answer, and with more than two choices in play a random guess is negative expected value against the skip.

• Checking only the first difference on a NumberLogic sequence. Several rule families here use a second-order recurrence or two interleaved subsequences, so a sequence that looks irregular on first differences may be perfectly regular split into two.

• Ranking Likelihood List items by intuition instead of computing the exact fractions - every one of the six possible orderings is offered as a choice, so a plausible-feeling wrong ranking is always available to pick.

• Leaving Zap-N or Zap-Q items blank under time pressure. Both are zero-penalty sections, so any answer - even an uncertain one - is strictly better than a skip.

Why interviews test this

The section split in this format is not arbitrary: each round is a scaled-down stand-in for a distinct trading competency. Raw arithmetic speed proxies for quoting a two-sided price before a counterparty walks; the sequence round proxies for spotting a repeating structure in noisy data; the probability and EV round is the closest of the six to the actual day-to-day job, since pricing a bet correctly and fast is what a market maker does all day; and the working-memory and pattern rounds proxy for tracking several open positions or signals without paper.

Firms use a battery like this, rather than a single test, because no one round reliably separates strong candidates from weak ones on its own - a candidate can be fast at arithmetic and still bad at updating on new information, or good at spotting patterns and still slow under a hard clock. Negative marking on three of the six sections is the other half of the filter: it is specifically designed to catch candidates who would guess through a live order book rather than quote only when they actually know the answer, which is the exact failure mode a trading desk cannot tolerate.