DSA

Learn why each algorithm works, then solve problems you have not seen.

One roadmap for interviews and contests: a lesson for every topic, and problems picked from Codeforces, AtCoder, CSES and LeetCode.

Lessons Each pattern explained, with traces you can step through Problem sets Side quests: Codeforces rating bands and USACO divisions Library Books, tier lists and problem sets worth your time Dan Ladder Pass each dan by solving its whole set Contests Upcoming rounds on Codeforces, AtCoder and LeetCode CP Guessr Guess a problem’s rating from its statement Quizzes Questions on complexity, data structures and algorithms

Premium

Open roadmap

The learning research behind WhiteBox

You get good at solving problems by solving them, even when reading solutions feels more productive.

Passive learning feels better. Active learning works better.

Most people watch the video solution, feel like they learned it, and make very little progress. The lecture students in this study felt they learned more and scored lower.

Ask Tsuki, the AI chat on every plan, for the minimal hint that unblocks you instead of the answer. That is scaffolding in Lev Vygotsky’s Zone of Proximal Development: just enough support to solve a problem slightly beyond what you can do alone.

Read: Is NeetCode / Strivers A2Z bad?

Active class versus a passive lecture

Active class versus a passive lectureActive class versus a passive lecture. Test of learning: +0.46. Feeling of learning: −0.56.−0.5 SDLecture+0.5 SDTest of learning+0.46Feeling of learning−0.56
149 students in introductory physics, taught the same material both ways. Deslauriers et al. (2019), PNAS

Get stuck before you read the editorial. You’ll understand it better.

You’ve probably been told to open the editorial after 20 to 30 minutes. Exhausting your own ideas first builds the reasoning.

The Core Catalog trains synthesis instead of recall. A problem combines several ideas instead of disguising one pattern, so you can’t finish it by matching it against something you memorized. You have to produce the insight yourself.

Read: Is pattern matching a trap?

Understanding of the concept, out of 10

Understanding of the concept, out of 10Understanding of the concept, out of 10. Taught first: 3.84. Tried the problem first: 6.33.0246810Taught first3.84Tried the problem first6.33
75 ninth graders learning standard deviation, tested right after the lesson. Kapur (2014), Cognitive Science

You can’t memorize your way through a follow-up question.

Studying one topic at a time means you always know which technique the problem needs, so you never face unseen problems, which are what an interview gives you. Students told the formula first kept up on ordinary word problems and fell behind on one that looked new.

Daily quests pay XP for solving a problem at your level on CSES, AtCoder, Codeforces or LeetCode, a different one each day, and White Room DSA deals problems you haven’t seen. Both hand you the problem without naming the technique that solves it.

Read: Meta-learning, and what Chinese OI gets right about learning algorithms

Score on a new kind of problem, three weeks later

Score on a new kind of problem, three weeks laterScore on a new kind of problem, three weeks later. Told the formula first: 23.0%. Invented it first: 53.5%.0%25%50%75%100%Told the formula first23.0%Invented it first53.5%
128 eighth graders learning density and speed in physics. Schwartz et al. (2011), Journal of Educational Psychology