IMC's online assessment is best known for one section: number sequences. Candidates describe a stream of series, each asking for the next term, on a clock tight enough that you either recognize the pattern or you do not answer at all. Where Optiver screens raw arithmetic, IMC's funnel leans on pattern recognition, and that difference should shape how you prepare.
The encouraging fact is that any sequences test draws from a small set of families. Learn the families, drill a fixed checking order, and most questions classify themselves within seconds. This guide gives you the taxonomy with one worked example per family, a decision tree for the first ten seconds, the traps, and a training method that actually transfers.
The Format, As Candidates Report It
Public candidate accounts describe a numerical screen built around number series: a run of terms, find the next one, answered by multiple choice or typed entry depending on the platform. The clock is the defining feature. Reports consistently emphasize that there is no time to derive a pattern from scratch on every question, only to recognize it.
Question counts, per-question timers, and marking schemes vary by region, role, and year, and IMC has run assessments through external platforms whose details shift between cycles. So treat any specific numbers you read with skepticism, and prepare for the skill rather than the wrapper.
The Pattern Families
Seven families cover the overwhelming majority of sequence questions in this style. One worked example each.
Arithmetic: constant difference
7, 12, 17, 22, 27, ?
Differences: 5, 5, 5, 5. Next term is 27 + 5 = 32. The fastest family to confirm, and always the first check. Watch for negative and fractional versions too: 9, 7.5, 6, 4.5 is the same family with a common difference of .
Geometric: constant ratio
2, 6, 18, 54, ?
Differences are 4, 12, 36. Not constant, but growing in proportion to the terms, and that proportional growth is the tell. Check ratios: each term is 3 times the last, so the next is 54 times 3 = 162. Halving runs like 160, 80, 40, 20 are the same family with ratio .
Second differences: the quadratic family
2, 3, 6, 11, 18, 27, ?
First differences: 1, 3, 5, 7, 9. Not constant, but climbing by 2 every step. When the second difference is constant, extend the first differences: the next one is 11, so the answer is 27 + 11 = 38. Any sequence whose differences form their own clean pattern lives in this family.
Interleaved strands: two sequences in one
4, 100, 7, 90, 10, 80, 13, ?
The differences sawtooth: , , , and so on. That jagged alternation is the tell. Split by position: odd positions run 4, 7, 10, 13, adding 3 each time, while even positions run 100, 90, 80, subtracting 10. The missing term belongs to the even strand: 70.
Recursive: terms built from previous terms
2, 5, 7, 12, 19, 31, ?
Each term is the sum of the two before it: 2 + 5 = 7, then 5 + 7 = 12, then 7 + 12 = 19, then 12 + 19 = 31. The next term is 19 + 31 = 50. A product variant exists as well: 2, 3, 6, 18, 108 multiplies consecutive pairs, so the next term is 18 times 108 = 1944. When a sequence accelerates faster than any geometric ratio explains, suspect recursion.
Digit-based: the digits drive the pattern
23, 28, 38, 49, 62, ?
Each term adds its own digit sum: 23 + (2 + 3) = 28, then 28 + (2 + 8) = 38, then 38 + (3 + 8) = 49, then 49 + (4 + 9) = 62. Next: 62 + (6 + 2) = 70. When the jumps look arbitrary but stay small, read the digits, not the numbers.
Anchors: primes, squares, cubes, with offsets
2, 3, 5, 7, 11, 13, ? is the primes, and the next is 17. No difference or ratio rule generates it, which is exactly the signature: when both standard checks fail and the numbers feel familiar, you are probably looking at a famous list.
Offsets disguise anchors. 2, 5, 10, 17, 26 is , so the next term is 37. Note that this one also falls to second differences, since the first differences are 3, 5, 7, 9, which is why the decision tree below catches many anchors without you ever naming them. Keep squares through , cubes through , and primes below 100 warm; they are the landmarks.
The First Ten Seconds: A Decision Tree
Run the same checks in the same order on every sequence. Speed comes from the order being fixed, not from inspiration.
- Differences first. Subtract neighbors left to right. Constant? Arithmetic, answer and move. Climbing steadily? Extend the differences and you have the quadratic family. Familiar in their own right, like 1, 3, 5 or 1, 2, 4, 8? Extend that pattern instead.
- Ratios second, and only if the differences grew in proportion to the terms. A constant ratio means geometric, done.
- Alternation third. If the differences sawtooth between positive and negative, or the magnitudes flip between two scales, split odd and even positions and read each strand alone.
- Anchors fourth. Terms hugging squares, cubes, primes, or powers of 2? Test the offset.
- Recursion last. Ask whether term three comes from terms one and two by sum or product, then verify on every later term.
Steps one through three resolve most questions. Every step also leans on fast subtraction and division, which is why sequence speed inherits directly from arithmetic speed. If neighbor differences cost you four seconds each, fix that first at the mental math trainer.
A Trap, Worked in Full
2, 4, 8, 14, 22, ?
The first three terms scream doubling: 2, 4, 8. A pattern matcher locks onto geometric, glances at the 22, and answers 44.
But the fourth term already broke that hypothesis. Doubling demands 16, and the sequence says 14. Run the routine instead: the differences are 2, 4, 6, 8, a clean arithmetic climb, so the next difference is 10 and the answer is 22 + 10 = 32.
The lesson generalizes. A hypothesis must survive every given term, not just the first three. Test writers know which prefixes are suggestive and choose them deliberately. The same trap shape appears as quadratics dressed in geometric clothing, near-primes dressed as add 2 then add 4, and interleaved strands dressed as noise. Confirming against all terms costs about a second and saves the question.
Pacing and Skip Strategy
Two rules hold regardless of what the marking scheme turns out to be.
First, cap your time per question. Recognition either fires within your cap or it will not fire at all under exam pressure. Where the platform allows revisiting, bank the fast ones, flag the rest, and return with whatever clock remains. Where it does not, the cap matters even more, because one stubborn question quietly eats three answerable ones.
Second, learn the marking before you walk in, because it sets your guessing policy. Reports on negative marking differ across IMC's platforms and cycles. If wrong answers cost marks, the standard skip math applies: guessing only pays above a confidence threshold, and for a plus one versus minus two scheme the breakeven is two thirds, a result the Optiver 80 in 8 guide derives in full. If there is no penalty, the policy inverts: a guess costs nothing and a blank earns nothing, so answer everything. Train both regimes so neither feels novel on the day.
How to Train
Memorizing question lists fails because the instances change. Recognition training works because the families do not.
The sequences simulator regenerates from the families above with fresh numbers on every run, under a clock, so the only thing you can learn from it is the thing that transfers: classification speed. Build it in three stages.
- Untimed classification. Spend a few days naming the family before solving anything. Accuracy comes first.
- Timed singles. Solve under a per-question clock and log every miss by family. Most people find one weak strand, often interleaved or digit-based, producing the bulk of their errors.
- Full runs. Simulated tests with two passes and deliberate skips. Watch your family-level accuracy in the results and drill the weakest family between runs.
Pair the sequence work with raw speed sessions at the mental math trainer, browse worked solutions in the question bank, and if you want the whole preparation sequenced for you, the structured tracks slot sequences practice alongside the rest of the timed test suite.
FAQ
How many questions are on the IMC sequences test?
Public reports differ by region, role, and year, and IMC has used different assessment platforms over time. What the accounts agree on is the pressure: little time per question, so recognition has to run at reflex speed. Prepare for the skill, not for one specific wrapper.
Is the test negatively marked?
Accounts differ by cycle and platform, so read your own invitation carefully. Train for both regimes anyway: with penalties, skip below your confidence breakeven; without them, answer every question. Switching policy is easy once the recognition itself is automatic.
Can you memorize past questions?
You can memorize lists, but you cannot make the test draw from them. Families repeat, instances do not. An hour of classification drills transfers to whatever numbers appear; a memorized list does not.
Drill It Until Classification Is Reflex
The gap between knowing these seven families and clearing the screen is repetition under a clock. The sequences simulator regenerates fresh patterns every run and scores you by family, so a weak strand has nowhere to hide. And once the test is behind you, the IMC firm page covers what the rest of the process looks like.
QuantPit is independent of IMC. Format details are as candidates publicly report them.