How it works

This is a work in progress. The early chapters are in; the rest of the book is being added over time, so a chapter you’re looking for may not have arrived yet.

The later a chapter was added, the more likely it still has rough edges — the problem text is transcribed from the book, and transcription slips. Reporting them genuinely helps, and there’s a link under every problem for it.

You set a status on each part; the problem’s own status is derived from them, first match wins:

• Every part solved → Solved
• Any part Needs review → the whole problem, even if the rest are solved
• Nothing started → Not started
• Anything else → Attempted

No partial credit anywhere — four of five parts solved is still just attempted. The dashboard counts and the random picker both work off this, so a problem with one unsolved part stays in the pool.

• $x^2$ — inline math. The closing $ is added for you, and it renders when the cursor leaves
• $$…$$ — a centered block. Press $ twice in an empty pair
• \ inside math — searches ~230 KaTeX commands
• \$ — a literal dollar sign, so prices stay prices
• Click a rendered formula to edit it; Esc reverts
• Select text and press $ to wrap it in math after the fact
• Tab hops out of a formula; 1. , - and a) start lists

It picks evenly among problems you haven’t solved. Solved ones are excluded outright, not merely made less likely — so you can lean on it until a chapter is finished and it will never hand you the same solved problem twice.

On the dashboard it stays within whichever chapter is selected. On the browser page it obeys whatever you’ve filtered to, which is the quickest way to drill one subtopic. On a problem page, hovering the button offers a choice between any chapter and the one you’re already in.

If everything matching is already solved, it falls back to that same set with solved problems allowed rather than sending you somewhere unrelated.

The browser page keeps its filters in the address bar, so a filtered view is a real link — bookmark it, or send it to someone and they see the same list. Individual problems have their own URLs too.

On the browser page, the line above the table that reads Showing N of M problems has a Recently viewed toggle next to it. That switches the table to the problems you’ve actually opened, newest first, and lets you choose the last 10, 20, or 50.

The filters dim while you’re in that mode — clicking any of them takes you back to the full list.

The dashboard also keeps a Continue Studying list of problems you’ve started but not finished.

Hover the Problem N heading for Copy problem and + notes. Both copy clean Markdown with $…$ math — rendered formulas turn back into their LaTeX source, so nothing is lost pasting into your own notes, an email, or an AI chat.

Hover Random Problem on a problem page to choose between any chapter and the current one.

On a touch device, tap instead of hovering.

Optional, and limited to Harvard Google accounts. Signed out, your work stays in this browser; signed in, it follows you anywhere. Sign in whenever — what you’ve already done is uploaded then.

Changes save to this browser the instant you make them, with no save button and no waiting — offline included. Signed in, they go up to the server a couple of seconds later in the background. If that fails, your work is still safe here and a quiet note appears in the account menu; it catches up when the connection does.

Notes are private to you.

There’s a Report an issue link under every problem. Chapter 2 was extracted from the book semi-automatically, so typos and formatting slips are expected and worth flagging.

Corrections to a problem belong on that problem, where the context is. Anything about the site itself — an idea, or something broken or confusing — goes here.