Drill Lab
Sunny Sums
An endless runner through a sunny field where the only way to hop is answering a mental-math question - one collision ends the run.
What is dual-tasking (divided attention)?
Dual-tasking is performing two attention-demanding activities at the same time rather than one after the other. Here that means solving an arithmetic problem and timing a motor action - the jump - concurrently, with the timing depending on the world (an approaching obstacle) rather than on when the calculation happens to finish.
It is a distinct skill from either task alone because attention is a limited resource, and psychology has long studied what happens when two tasks compete for it: even activities that are trivially easy in isolation measurably slow down, or become error-prone, once something else is also claiming attention. The added difficulty in a dual-task setting is not the arithmetic itself getting harder - it is the same arithmetic, done while something else needs watching.
What makes dual-tasking trainable is that fluency in one of the two tasks frees attention for the other. This is the same mechanism behind a practiced typist being able to talk while typing, or an experienced driver holding a conversation on a familiar route: once a component task stops requiring conscious effort, it stops competing for the attention the other task needs.
Trading desks are dual-task environments almost by definition - a pricing decision, live market data, risk limits, and communication all compete for attention at once. Interview exercises that pair arithmetic with a second, concurrent demand are testing something a pure math quiz cannot: whether a candidate's reasoning quality holds up when attention is split, not just when the problem has the room's full, undivided focus.
How it works
Sunny Sums is an endless runner, with one rule that is the whole game: the space bar submits your typed answer to the current math question, and a correct answer IS the hop - fired at that instant, or not at all if you are already airborne. There is no stored credit and no auto-timing; you are solving the question AND timing the press so the hop clears whatever is coming. Pressing early and landing before it arrives spends the answer for nothing, and a wrong answer just shakes the input (the box clears for a clean retry) while the obstacle keeps coming.
Questions come from a ten-tier difficulty ladder, ten questions per tier, mixing arithmetic, fractions, and percentages - you can also restrict the run to one of those categories from the menu. Tier 1 is deliberately trivial single-digit work and the ladder climbs to genuinely hard tiers 8 through 10. Every answer is a whole number or a fraction already in lowest terms, and grading is exact string match, so a fraction answer must be typed reduced.
You run as your own avatar - whichever one you picked on your profile, or the duck if you have not - and hop flowers, thick patches of grass and apples. The run ends on the first collision: one hit, no lives, no second chance. Speed ramps with your score, and the space bar also starts and restarts runs from the menu and game-over screens.
How scoring works
Score is simply the number of questions answered correctly - each correct answer scores 1 and deals the next question. Wrong answers cost nothing beyond the obstacle still bearing down.
Difficulty is adaptive rather than score-marched: a controller watches whether you answer correctly and how fast (the fluency gate is 9 seconds per question, sized for doing two things at once), promoting fluent players up the ten tiers and easing struggling players down. Your tier is saved locally and resumes next session instead of restarting the climb.
Game speed ramps linearly with score from 3.6 to a cap of 12, pinned so maximum pace lands exactly when the hardest tier's questions do (score 90). The game-over screen shows an estimated percentile for your score.
Solve ahead, jump on time
The core skill is decoupling the two halves. Solve the question the moment it appears and type the answer immediately - but hold the space bar until the obstacle is close. Your typed answer sits in the box costing nothing; the press is the only timed action. Players who treat answer-and-press as one motion burn their jump early, land in front of the obstacle, and die with a correct answer already spent.
The jump has fixed physics - same arc every time - so the timing is learnable as a pure motor skill: press when the obstacle reaches a consistent landmark distance from your runner. Spend your first few runs consciously calibrating that landmark at low speed, because the same landmark compresses in time as speed ramps and your press has to come at the landmark, not at a remembered rhythm.
If you are still solving when an obstacle gets close, that is the one situation to hurry - a fast correct answer late still jumps, a perfect answer after the collision does not. This is where the early tiers' trivial questions matter: bank them instantly so you are never solving under the obstacle.
Keep answers in the exact expected form
Grading is exact string match against an answer key that is always a whole number or a fully reduced fraction. Percentage questions are built so the answer is a whole number - if your percentage arithmetic produces a decimal, you have made an error, not found a decimal answer. Fraction questions like 1/8 + 3/8 want 4/8 reduced to 1/2. Typing an unreduced or decimal form of the right value counts as wrong.
Percentages here reward the fraction-conversion table: 25 percent is a quarter, 12.5 percent an eighth, 20 percent a fifth, 75 percent three quarters. "75% of 48" as 48 x 3/4 = 36 is two beats; long-form percentage multiplication is five. The bases are generated to divide cleanly, so the fraction route always lands on the whole number the key expects.
For percent-change questions ("140 increased by 15%"), compute the delta and add: 10 percent of 140 is 14, 5 percent is 7, so +21 gives 161. Splitting a percentage into 10s and 5s and 1s is the fastest reliable route and keeps everything in whole numbers.
Ride the adaptive controller deliberately
The tier you play is not a reward ladder, it is a fluency estimate: answer correctly inside about 9 seconds and you drift up; miss or stall and you drift down. That means accuracy-plus-tempo is what promotes you - firing fast wrong guesses is doubly bad, dropping your tier and teaching nothing. A steady sub-5-second correct pace climbs the ladder quickly.
Because your tier persists across sessions, warm-up runs matter less than they would in a reset-every-time game: you resume where your fluency left you. Use the category filter to fix specific weaknesses - a fractions-only or percentages-only session drills the exact conversion habits the mixed ladder will later demand at the high tiers.
Late runs fail at the intersection of hardest questions and maximum speed, by design - both ramps are pinned to the same score. Past score 90 the game stops getting faster, so the survival problem becomes constant and the only variable left is how fast you clear tier-10 questions. That is the regime where the fraction and percentage shortcuts stop being nice-to-have and become the run.
A worked example
The prompt: 3/4 of 32. Use the fraction route instead of long multiplication.
Divide by the denominator first.
Multiply by the numerator for the answer.
Type 24 and hold - the answer sits ready while only the jump's timing remains, decoupling solving from pressing space.
Press space when the obstacle reaches the calibrated landmark, not the instant the answer is ready - the jump fires at that press.
Common mistakes
• Pressing space the instant you finish solving - an early jump lands before the obstacle arrives and the correct answer is spent for nothing.
• Submitting unreduced fractions or decimal forms - grading is exact string match, so 4/8 and 0.5 both fail where 1/2 passes.
• Doing percentages long-form instead of converting to fractions - the bases always divide cleanly, so 75% of 48 is 48 x 3/4.
• Firing fast guesses under pressure - misses ease the adaptive tier down and the obstacle is unmoved either way.
• Timing jumps by rhythm instead of by the obstacle's position - the speed ramp compresses the rhythm but not the landmark.
Why interviews test this
Doing arithmetic while managing a second real-time task is closer to a trading desk than any pure math quiz: the calculation is easy, the constraint is doing it while something else demands attention on its own schedule. Timed mental-math screens in trading recruitment exist to find people whose arithmetic survives divided attention.
The one-hit format also mirrors the asymmetry the culture drills in: in a fast market, one unforced error can end an otherwise good day. Playing for a long steady run over a flashy fast one is the same temperament interviewers look for.