Clarify first: questions, assumptions and the rules list

A problem statement in a coding round is short, often three or four lines. The real problem is longer. The statement hides many rules, and the hidden tests check them. In lesson 7 you saw that clarifying is worth 10 of 100 points. It also protects the 25 points for working code. If you guess a rule wrong, the tests fail, and you only find out at the end.

A weak candidate starts to type at once, or asks 20 questions in a row, or asks what the statement already says. A strong candidate asks 5 to 8 questions, says the assumptions aloud, and writes a numbered list of rules. Then the candidate codes from that list.

At SDE-1 level you should ask about the rules. At SDE-2 level you should also ask about bad input and about what is out of scope. At Senior level you should also name your own assumptions. These cover the size of the data, many callers at the same time, and where the data lives. For each one, propose a default.

Think about ordering a shirt from a tailor. You say: "Make me a shirt." A good tailor does not start to cut. The tailor asks about the size, the collar, the colour and the date, and writes it all on a slip. Later, when you say "that is not what I asked", the slip shows who is right.

The problem statement is your "make me a shirt" sentence. A clarifying question is a question that makes one hidden rule visible. A rules list is the slip: a short numbered list of what the code must do, with one example for each rule. In lesson 3 you learned to ask for requirements before a design. This is the same habit for code, with smaller questions.

Seven groups of questions

Use these seven groups as a checklist. You do not ask all of them every time. The examples come from a problem we use below: coupon codes for a food-ordering app.

1. Who uses it and what do they do. "Does one order use one code, or can it use two?"
2. Rules and numbers. Limits, prices, rounding and ties. A tie is two values that are equal, so the rule cannot pick one. A cap is the most that something can be, for example a discount of at most Rs 50. "Is the discount a percent or a fixed amount? Is there a cap? Do we round the discount up or down?"
3. Bad input. A bad input is a value the code should not accept, such as a code that does not exist. "What do I return? Does anything change when the input is bad?"
4. Time, ids and order. "How does today's date reach the code? What do the ids look like? Is the order shown as text order or number order?" Text order puts C10 before C9, because the text 1 comes before 9.
5. Size and many callers. "How many items? Does one caller use the code, or many at the same time? Is everything kept in memory only?"
6. Scope. What is out of scope: left out on purpose. "Do I build payment, saving data on disk, or a screen?" The answer is usually no. Ask so that you do not build it.
7. A sample run. "Can you give me one example with real numbers?" It tells you the format of the input and of the answer.

Say an assumption aloud

Sometimes the interviewer says: "Your choice." Do not guess in silence. An assumption is an answer that you choose and say aloud because you do not know the real one. Use one fixed sentence: "I assume X. Tell me if not." The interviewer can agree in two seconds or correct you. Both are good for you.

Words that hide rules

Some words in a statement look clear but hide a rule. Mark them and ask.

Look at every boundary. A boundary is the exact number where a rule changes. Take "at least Rs 500". Ask about the three cases: below 500, exactly 500, and above 500. The middle case is the one that people forget.

Turn the answers into a rules list

When you can write 3 sample calls, stop asking. Then write the rules list:

• Number every rule. The numbers let you say "rule 4" during the round.
• Write one rule in one sentence.
• Put one small example next to every rule that has a number in it.
• Write the assumptions in the list too, marked with the word "assume".
• Write the answers down. Do not keep them in your head, because you will need them again later in the round.

Take 5 to 10 minutes in total. In the practice round of lesson 7 the limit for the whole clarify step is 10 minutes. Five to eight questions fit in that time.

The statement is: "Build coupon codes for a food-ordering app. A customer applies a code to an order and gets a lower total." That is all. Here are eight questions and the answers an interviewer gives.

Question 4 hides a point about order. The date 2026-11-10 is year, month, day. Text of this shape sorts correctly as text, because the biggest part comes first. The date 10-11-2026 would not. Question 3 matters too. Ten percent of Rs 205 is 20.5, and the answer says 20.

The rules list. Assumption: two codes exist.

• SAVE10: 10 percent off, cap Rs 50, minimum Rs 200, valid till 2026-11-30.
• FLAT100: Rs 100 off, minimum Rs 500, valid till 2026-11-20.

The rules are:

1. A code is percent or flat. Percent discount = order total x percent / 100, rounded down, and at most the cap. Flat discount = the fixed amount. Example: SAVE10 on Rs 560 gives 560 x 10 / 100 = 56, the cap is 50, so the discount is Rs 50.
2. The order total must be at least the minimum. A total equal to the minimum works. Example: FLAT100 on Rs 499 is refused, on Rs 500 it gives Rs 400.
3. A code works up to and including its last day. Example: SAVE10 works on 2026-11-30 and is refused on 2026-12-01.
4. A customer can use a code once. A refused call does not count as a use. Example: a second SAVE10 call by u1 is refused.
5. One code for each order. Example: SAVE10 and FLAT100 on the same order is not allowed.
6. Unknown code, expired code, small order, used code: refuse with a reason, and change nothing. Example: NEWYEAR is refused as unknown.
7. Everything stays in memory, with one caller at a time. Assume: amounts are whole rupees, and the new total is never below 0. Example: SAVE10 on Rs 205 gives 205 x 10 / 100 = 20.5, rounded down to 20, so the total is Rs 185.

Three sample calls. These lines are plain text, not code to run.

apply("SAVE10",  "u1", 560, "2026-11-10")  ->  510
apply("SAVE10",  "u1", 560, "2026-11-10")  ->  refused: already used
apply("FLAT100", "u2", 499, "2026-11-10")  ->  refused: below the minimum of 500

The calculation for call 1. Ten percent of Rs 560 is 560 x 10 / 100 = Rs 56. The cap is Rs 50, so the discount is the smaller number, Rs 50. The new total is 560 - 50 = Rs 510.

The predict box. "At least Rs 500" is a boundary, so you ask a question from the group "rules and numbers": "Does Rs 500 itself count?" The interviewer said yes in question 3. Your list says so in rule 2, with an example. FLAT100 on Rs 499 is refused. On Rs 500 it gives 500 - 100 = Rs 400. On Rs 501 it gives Rs 401. If the interviewer had said "your choice", you would say: "I assume Rs 500 counts. Tell me if not."

You asked 8 questions, wrote 7 rules and 3 sample calls. Assumption: each question and answer takes about 40 seconds. Then the questions take 8 x 40 = 320 seconds, which is about 5 minutes. That leaves 10 - 5 = 5 minutes to write the list and the sample calls.

Asking first solves the risk of building the wrong rules. It costs minutes that you could use for code. If you ask too many questions, or ask what the statement already says, the interviewer may think you cannot decide. I would shorten the questions when the round is short, say the assumptions aloud in one sentence each, and start coding sooner. I would ask more when the statement is vague and the cost of a wrong guess is a rewrite.

"The statement is complete, so a question looks weak." The statement is short on purpose. Interviewers expect questions, and Tradeoff gives 10 points for them. A question that shows a hidden rule (the boundary, the rounding) looks strong.

"The more questions I ask, the better." Twenty questions use half the clarify time and show that you cannot decide. Ask 5 to 8, and for the rest say an assumption.

"An assumption is a guess, so it is risky." A silent guess is risky. A spoken assumption is safe: the interviewer can correct it at once, and it goes into your rules list.

1. Name the seven groups of questions.
2. Using the rules list above, a customer applies SAVE10 to an order of Rs 205 on 2026-11-10. What is the new total? Show the discount step by step.
3. A statement says: "Give the nearest driver a free ride if the trip is at least 5 km." You have two minutes. Which words hide rules, and what do you ask or assume?