First Principles Thinking

Break a business problem down to what is actually true, then build your answer back up. Get coached on how well you reasoned.

Industry

Function

Difficulty

Your reasoning

Attempt 1

Assumptions I'm challenging

What does everyone take for granted here? Which "rules" might not be true?

Fundamental truths

Strip it to basics: unit economics, raw input costs, the customer's real job, hard constraints. Use the numbers.

My solution, built up from those truths

Reason step by step from step 2. Say what you'd do and why it follows.

CHALLENGE A FRIEND · SAME CASE · SAME CLOCK

Who thinks it through better?

Pick a business case (15 minutes) or a Logic Sprint (5 questions, 90 seconds each).
Your friend gets exactly the same one. Start together, live, or each on your own clock within 24 hours of them accepting.
Case: the higher score wins, and scores within 3 points are a draw. Sprint: more correct answers wins, and a tie goes to the faster time.
Neither of you sees the other's score until you've both finished. Your result also counts towards your own rating.

Topic

Difficulty

Your pseudocode

Plain-English steps or any programming language

Pseudocode rules: what one step may do

Allowed in one step

Set a variable, do arithmetic (+ − × ÷, remainder) and compare values
if / else if / else
Loops: for each item in a list, for i from a to b, while a condition holds
Lists: make an empty list, add an item, get the item at a position, length, sort (by a value, up or down), reverse
Maps / dictionaries: make an empty map, set the value for a key, get a value (with a default), check if a key exists, list the keys
Text: length, character at a position, split, join, upper/lower case, compare
min, max or sum of a list of numbers
Output or return a value

Not allowed (counted as missing logic)

A single step that solves the task or a large part of it ("find the answer", "group by customer and pick the top one", "find the shortest route")
Calling a ready-made library function that does the core job
"etc.", "…" or "and so on" in place of logic

Example of a good step: for each (name, amount) in entries: add amount to totals[name]. Not allowed: find the customer who owes the most.

Number or indent your steps, name your variables, and say what you output for empty or tie cases. The examples are public; your answer is checked against 7 hidden tests.

How you're scored

Correctness 8 points for each of 5 hidden normal tests your logic passes	40
Edge cases 10 points for each of 2 hidden edge tests (empty input, extremes, ties)	20
Efficiency Meets the target complexity: 15. One step slower: 8. Slower: 0. Starter tasks have no target.	15
Clarity 0, 5, 10 or 15 for named steps, clear variables, indentation, no ambiguity	15
Completeness 0, 5 or 10 for covering every input and output the task asks for	10
Total	100

The AI traces your steps against each test exactly as written. It doesn't fix your bugs. The points are then added up by fixed rules.

Your tasks

Your scores appear here after your first graded task.

What is first principles thinking?

First principles thinking means breaking a problem down to what is actually true, such as costs, constraints and what customers really need, then building your answer back up from those truths instead of copying what others do. It is how strong product managers, founders and consultants find options everyone else misses.

First Principles Thinking gives you a fresh business case in the industry you choose, asks you to challenge its assumptions, reduce it to fundamentals and rebuild a solution, then scores your reasoning out of 100 with specific feedback. You can read every case in Hindi, Bengali, Marathi, Telugu, Tamil, Gujarati, Kannada, Malayalam, Odia, Punjabi or Urdu, and answer by voice in your mother tongue. Idea Lab pressure-tests product ideas with a tough AI investor, and Logic Sprint is a timed, five-question test of data tables, deduction and business reasoning. Read the guide to first principles thinking.

The AI coach scores how you reasoned, not whether you reached a particular answer. Copying what competitors do scores low. Breaking the problem into basic truths and rebuilding from them scores high.

Case Practice · out of 100

Your answer is marked on 5 skills, each from 1 to 5. Your score is the total × 4, so all 5s is 100 and all 3s is 60.

Challenged assumptions	Did you name and question what everyone takes for granted, including the obvious answer?
Reduced to fundamentals	Did you get down to basic truths (costs, customer needs, hard limits) instead of "others do X"?
Used the evidence	Did you work with the numbers in the case?
Rebuilt a solution	Does your answer follow step by step from your fundamentals?
Original vs conventional	Did you go beyond the usual answer, for a good reason?

85–100 First-principles fluent70–84 Thinking from fundamentals40–69 Getting to the rootBelow 40 Reasoning by analogy

A blank or vague step scores 1 on the skills it covers. Revising usually lifts your score, and the coach tells you what improved.

Idea Lab · out of 10

The investor scores 8 areas from 1 to 5. Problem and Fundamentals count 20% each; Spark, First Step, Payoff, Stake, Frequency and Responsibility count 10% each. The weighted result is turned into a score from 0 to 10.

Strong habit potential: 7+ with a solid habit loop and real frequencyUseful but occasional: 4.5+Rethink the core: below 4.5

Areas the interview never reached, or answered vaguely, score 1 or 2. Specific answers with real numbers and situations score higher.

Logic Sprint · out of 100

5 multiple-choice questions on data tables, logical deduction and business reasoning (at least one of each), 20 points each. Wrong or unanswered answers score 0; there's no negative marking.

Time	90 seconds per question. When it runs out, the sprint moves on and that question counts as unanswered. You can't go back.
Leaving the page	Switching tabs or apps, minimising or locking your phone submits the sprint straight away. If the window just loses focus, you get 3 seconds to come back.
Checked answers	The AI writes 7 questions with an answer key. A second, independent AI pass solves them without seeing the key, and only the questions where both agree are used.
Rating	Logic Sprint has its own rating line and moves gently: K = 16 instead of 40. Sprints that go over the time limit don't count.

Coders Bay · out of 100

The AI traces your pseudocode, step by step and exactly as written, against 7 hidden tests, then the points are added up by fixed rules:

Correctness · 40	8 points for each of 5 normal tests your logic passes.
Edge cases · 20	10 points for each of 2 edge tests: empty input, extremes, ties.
Efficiency · 15	Your approach's time complexity against the task's target: meets it 15, one step slower (e.g. O(n log n) instead of O(n)) 8, slower 0. Starter tasks have no target.
Clarity · 15	0, 5, 10 or 15: named steps, meaningful variable names, consistent indentation, no ambiguous steps.
Completeness · 10	0, 5 or 10: every input and output the task specifies is handled, including empty or "none" cases.

Pseudocode rules: each step may only use simple building blocks (variables, if/else, loops, list and map operations, min/max/sum, output). A step that solves the task in one go, like "find the customer who owes the most", counts as missing logic, so the tests that depend on it fail. The full list is above the answer box.

Checked tests: every hidden test's expected answer is verified by running a reference solution before you see the task. Tasks whose tests don't check out are thrown away.

After grading you see each test, whether you passed it, and why, plus one clean solution. Any programming language is accepted as well as plain-English steps.

Challenge a friend

You both get the same case or sprint, made before either of you sees it. A case gives you 15 minutes; a sprint is 5 questions of 90 seconds each. The clock runs on the server, so refreshing doesn't add time.

Case: each answer is scored out of 100 as usual. The higher score wins, but scores within 3 points are a draw, because AI marking can vary by a few points. Sprint: more correct answers wins; if that's equal, the faster total time wins.

You see your friend's score only after you've finished. If one of you doesn't play before the challenge closes (24 hours after it's accepted), the other wins. Your own result also counts towards your rating, like any first attempt.

Your rating · works like chess

You have a separate rating for Case Practice, Idea Lab and Logic Sprint. Everyone starts at 1000. Each result is compared with what's expected at your rating for that difficulty:

Rating change = 40 × (your score as a fraction − expected fraction), where expected = 1 ÷ (1 + 10^{(difficulty − your rating) ÷ 400}), the same formula chess ratings use. Difficulty counts as Starter 900, Intermediate 1100, Hard 1300, and Idea Lab 1000. Logic Sprint uses 16 instead of 40, so each quick sprint moves it gently.

Rating 1000, Hard case	Expected about 15/100. Score 60 → +18.
Rating 1000, Intermediate case	Expected about 36/100. Score 60 → +10.
Rating 1000, Starter case	Expected about 64/100. Score 60 → −2.
Rating 1000, Idea Lab	Expected 5.0/10. Score 6.5 → +6.

As your rating rises, the same score earns less, so harder cases become the way to climb. Only your first graded submission on each case or task counts; revisions still improve your score. Sample cases and the sample task are for practice and don't count. Every finished Idea Lab verdict counts.

Scores come from an AI, so the same answer can land a few points differently on another try. Use them to see where your thinking is strong and where to push, not as a certificate.

First Principles Thinking

Your reasoning

Pitch your idea. Survive the grilling.

The investor is listening

How sharp is your logic today?

Who thinks it through better?

Your reasoning

Turn your training material into a spoken training

Your pseudocode

Allowed in one step

Not allowed (counted as missing logic)

Why train with us?

What is first principles thinking?

Your cases

Case Practice · out of 100

Idea Lab · out of 10

Logic Sprint · out of 100

Coders Bay · out of 100

Challenge a friend

Your rating · works like chess

Each correct answer	20
Wrong or unanswered	0
Total (5 questions)	100