Jane Street Probability Playing Cards: Which Problems Should You Practice for Interviews?

For interview preparation, start with Jane Street's own recommendation: the first five or six questions in each of clubs and diamonds. Use those questions to check foundations and practise explaining your reasoning. Completing the hardest cards is not the target Jane Street sets for this purpose.

This article distinguishes official deck guidance from our preparation recommendations and original exercises. It does not reproduce the cards, their solutions, or a claimed bank of actual interview questions. No candidate reports are used to establish question frequency or a current hiring sequence. The scope is selecting the deck's preparation subset and identifying what to practise when a clubs or diamonds problem reveals a gap.

Treat adjacent practice as an opportunity to explain a method, not evidence that a particular question will appear at Jane Street.

Which cards does Jane Street recommend?

Official fact: The Probability Playing Cards page assigns expectation and linearity to clubs, classical and conditional probability to diamonds, geometric probability to hearts, and random walks and martingales to spades. Difficulty rises roughly from ace to king. A single asterisk marks a very hard problem; two indicate an unsolved one. The page recommends the first five or six questions in clubs and diamonds for interviews and describes most of the deck as harder than what is needed to work there.

That recommendation is a starting range, not a readiness score. Finishing it does not establish that you can handle a changed assumption, explain your result aloud, or solve an unfamiliar problem. Conversely, getting stuck on a recreationally difficult card does not by itself establish a gap in interview-level foundations.

Our preparation recommendation is to let the recommended subset reveal a specific weakness. If a clubs problem exposes confusion about an expected count, work on indicators. If a diamonds problem exposes confusion about conditioning, identify exactly which outcomes remain possible after the observation. Choose the next exercise from that diagnosis.

The original decision map below turns that approach into an action:

What happened during your attempt?	Skill to revisit	Useful next evidence
You can enumerate cases but cannot generalize an expected count	Indicator variables and linearity	Derive the count for a changed sequence length.
You add event probabilities without checking overlap	Counts versus “at least one” events	Explain what happens when two counted events occur together.
You condition on an observation without defining it	Sample space and selection mechanism	List the surviving outcomes under the stated observation.
You recognize a trick but cannot explain why it applies	Modeling and assumptions	Solve a small variant without using the remembered conclusion.
You reach the right result but your explanation is hard to follow	Communication and checking	Give the model, method, result, and one independent check aloud.

Hearts and spades can be interesting extensions when they match your learning goals. Our recommendation is to avoid using a difficult extension as an escape from an unresolved basic error. If you confused a count with a probability, another short indicator exercise gives more targeted feedback than an unrelated difficult puzzle.

What if the recommended cards still feel unfamiliar?

Official fact: Jane Street's Probability & Markets page introduces a guide covering probability, expected value, and making markets for quantitative trading and research preparation. The page explicitly treats it as optional reading rather than a required qualification.

Use that official starting point when vocabulary is the obstacle. Then check whether you can define an event, distinguish it from a random variable, and calculate a small expectation from its possible values. These are our prerequisite checks, not a published Jane Street entrance test.

For conditional probability, state what was observed before writing a fraction. “The final toss was heads,” “at least one toss was heads,” and “a randomly chosen heads was shown to you” can describe different information. A solution needs the observation rule, not only a symbol that looks like Bayes' formula.

For expectation, identify what is being counted. A variable can count several successes in one outcome, whereas an event indicator is either zero or one. This distinction becomes important when successes overlap. The following exercise isolates that issue without requiring an advanced theorem.

Original exercise: two overlapping opportunities for HH

The exercise in this section is original interview practice, not a reproduced deck card or candidate-reported interview question. Toss three fair, independent coins. Let X count adjacent pairs that are both heads: the pair at positions 1–2 and the pair at positions 2–3.

Before calculating, clarify that an outcome such as HHH contains two qualifying adjacent pairs. The question counts windows; it does not count disjoint groups or ask merely whether HH appears somewhere.

Define I1 as one when the first two tosses are HH, and zero otherwise. Define I2 similarly for the last two tosses. Then X = I1 + I2. Each pair has probability 1/4 of being HH, so each indicator has expectation 1/4. Therefore:

E[X] = E[I1] + E[I2] = 1/4 + 1/4 = 1/2

The coin tosses are independent, but the two pair events are dependent because they share the middle toss. Indeed, both pairs are HH only when the outcome is HHH, with probability 1/8, whereas multiplying their marginal probabilities gives 1/16. Linearity of expectation does not require the indicators to be independent.

An exact enumeration provides an independent check:

Outcome	Number of HH windows
HHH	2
HHT	1
HTH	0
HTT	0
THH	1
THT	0
TTH	0
TTT	0

All eight outcomes have equal probability under the stated model. Their counts total four, so the average is 4/8 = 1/2. This confirms the indicator calculation without relying on an independence assumption between windows.

Now answer a different question: what is the probability that at least one HH window occurs? The favorable outcomes are HHH, HHT, and THH, giving 3/8. Expected count 1/2 and occurrence probability 3/8 are different because HHH contributes two to the count but only one to the occurrence indicator.

In an explanation, make that distinction before reaching for more formulas. If you simply report “the probability is one-half,” ask which random quantity you actually calculated. A correct expectation becomes an incorrect answer when it is attached to the wrong question.

Change the observation, then rebuild the sample space

Keep the same original exercise, but suppose you are told that the final toss is heads. This means conditioning on a particular position's result, not hearing an unspecified report that some heads occurred.

The surviving outcomes are HHH, HTH, THH, and TTH. They remain equally likely under this observation, and their HH counts are 2, 0, 1, and 0. Thus E[X | final H] = 3/4. Two of these four outcomes contain at least one HH, giving conditional occurrence probability 1/2.

You can also check the conditional expectation with indicators. Given final H, the first pair is HH with probability 1/4, while the second pair is HH exactly when the middle toss is H, with probability 1/2. Their expectations sum to 3/4. Again, independence between the two indicators is unnecessary.

If the observation changes to “there was at least one heads anywhere,” the denominator changes. Seven outcomes survive, and the counts still total four, giving expected count 4/7; occurrence probability becomes 3/7. You cannot keep the final-position denominator and simply rename the condition.

Translate the observation into surviving outcomes and their conditional weights. Equal weighting is justified here by independent fair tosses and a deterministic event filter. If the problem instead describes someone selecting an outcome to show you, ask how that selection works before assuming the same weights.

Generalize the count or change the stopping rule?

For n independent fair tosses, with n at least two, there are n−1 adjacent windows. Each is HH with probability 1/4. The expected number of HH windows is therefore (n−1)/4. Adjacent windows still overlap, and linearity still applies.

Do not use that expression as an occurrence probability. At n = 6 it equals 5/4, which is a valid expected count and an impossible probability. This simple boundary check exposes the interpretation error immediately.

A different original extension asks how many tosses it takes to see HH for the first time. The fixed-length count formula does not answer this stopping-time question. Track whether the most recent toss supplies one useful H toward the pattern.

Let E0 be the expected additional tosses from no useful trailing H, and E1 from one trailing H. A tails returns you to state 0; a heads from state 1 finishes the pattern. Under fair independent tosses:

E0 = 1 + (E0 + E1)/2
E1 = 1 + E0/2
E0 = 6; E1 = 4

The leading one accounts for the next toss. Explain both transitions rather than memorizing six. This extension checks whether you can adapt your model when the question changes from a count within a fixed sequence to waiting for a pattern. This extension is our practice suggestion, not an additional requirement published by Jane Street.

How should you review an attempt?

Official fact: Jane Street's trading interview guidance emphasizes methodical work, clear communication, correcting mistakes, and asking why. It describes trading interviews as conversational and collaborative. That guidance is role-scoped; this article does not infer an identical process for every Jane Street position.

Our practice recommendation is to replay one mistake as a changed assumption. Instead of recording only “wrong answer,” use the original exercise to name the actual failure and the next check:

Error in the HH exercise	What the error means	Next check
Claiming the windows are independent	Shared toss was ignored	Compare 1/8 joint probability with 1/16 product.
Calling 1/2 the occurrence probability	Count and event were confused	Explain HHH's contribution to both quantities.
Reusing four outcomes after a different observation	Conditioning rule was not rebuilt	List the outcomes that survive the new event.
Recalling six without the state transitions	Stopping-time model is missing	Explain E0 and E1 before solving them.

Have a practice partner ask one follow-up, then correct the model aloud if necessary. Identify what changed and which part of the old solution remains valid. Explain the correction and the assumption that caused the error.

No deck completion percentage, private practice timer, or solved-card count is a validated hiring threshold here. If you need role-specific preparation beyond probability, consult the role's actual guidance and your invitation rather than extrapolating an entire assessment from this deck.

Additional probability practice

Question	Skill to transfer from the deck practice
Expectation and Variance of a Sum of Two Random Variables, Including Covariance	Distinguish expectation's linearity from variance's dependence terms.
Calculate Expected Streaks in Coin Toss Sequence	Define the counted windows and their overlap.
Four Fair Coins: Expected Payoff and Pricing Two Magic Wands	State the payoff variable before calculating its expectation.
Condition on a Chosen Rabbit Losing	Specify the observation and selection mechanism.
Compute Commuter Ride Probabilities	Check event definitions and conditional weights.

Start with one Statistics & Math question that targets the weakness you identified. Write the assumptions, explain the result, and change one condition. Use the result to choose your next exercise; it does not establish an interview-readiness score.

Sources and Further Reading

• Jane Street: Probability Playing Cards conventions and preparation subset
• Jane Street: optional Probability & Markets preparation guide
• Jane Street: trading interview guidance and mock interview