Two 50-minute blocks on a weekday, a real break between them, and one full day a week at zero. That is the ceiling, not the target, and holding it as a ceiling is the entire skill. Past roughly two focused hours the problems still get solved, but they stop getting learned, and the bill arrives three weeks later as a bookmark you cannot make yourself click. Around eight hours of practice a week, run this way, clears a standard coding screen.
Why the third hour costs more than it returns
What makes a problem stick is retrieval under strain. You read a statement cold, sit in the confusion, name a family, get it wrong, and fix it. That is expensive in a way that reading is not, and the tank for it is small: two blocks for most people, sometimes three on a clear day with nothing else going on.
After the tank empties you do not stop working. The work changes shape instead. You skim the statement rather than restating it. You open the editorial at minute eight instead of minute twenty-five. You type a template you half remember and let the judge tell you whether it was right. The session logs three solves and encodes none of them, which would merely be a waste, except that it also charges interest: an hour spent in that state lowers your willingness to start the next one. That is the real mechanism behind LeetCode burnout. Not effort, but hours of low-quality effort that felt like effort.
| Daily load | How weeks one and two feel | Where it usually breaks | When it is the right call |
|---|---|---|---|
| One 50-minute block About 4 hrs/week |
Slow. You will feel behind, and you are not. | Rarely breaks. The risk is the calendar, not you. | A three-month horizon, a heavy sprint at work, or the week you come back from a break. |
| Two 50-minute blocks About 8 hrs/week, one zero day |
Productive and slightly unfinished, which is the feeling you want. | Holds for months, provided the zero day is real. | The default for anyone prepping alongside a job or a degree. |
| Three to four hours Daily, no fixed rest |
Excellent for about ten days. Big numbers, real momentum. | Week three to five. Editorials creep earlier, then one skipped day becomes a skipped fortnight. | A bounded sprint with a known end date, split by a long midday gap, with mocks and a zero day still non-negotiable. |
| Six-plus hour grind days | Two or three of them feel heroic. | Usually inside a fortnight, and it takes the habit down with it. | Almost never. If the onsite is that close, those hours belong in mocks, not new problems. |
The two-block weekday, minute by minute
Structure matters more than the total. Two vague hours with a phone face-up are worth less than one hundred minutes run like this.
Block one, 50 minutes, one new problem
Minutes 0 to 5: restate the problem in your own words on paper, then write down the family you think it is and the complexity you are targeting. Minutes 5 to 25: attempt it with nothing else open. At minute 25, if you have no working approach, take exactly one hint, never the full solution. Minutes 25 to 45: code it and hand-trace two examples before you submit. Minutes 45 to 50: write three lines, the family, the cue that should have given it away, and the mistake you made.
A break that is actually a break, 10 to 15 minutes
Stand up and leave the screen. Not the discussion tab, not a solution thread, not a scroll through anything. If you spend the break reading somebody else's approach, you did not take a break, you extended block one at lower quality, and you will feel it in block two. Getting out of the chair does more here than the exact number of minutes.
Block two, 50 minutes, and it is not new problems
Open a blank file and re-solve one problem you did seven to ten days ago, from memory, notes closed until you are done. This is the block that converts exposure into recall, and it is the one everybody skips. If the re-solve takes under 20 minutes and felt clean, spend what is left on one fresh medium. If it did not, spend the rest on a second problem from that same family and stop there.
Hard stop, even mid-problem
Close the tab when block two ends, including on a problem you are halfway through. Write one sentence about where you got to and what you would try next. An unfinished problem waiting for you is the cheapest possible start to tomorrow, and stopping while it is still going well is the most reliable way to keep wanting to come back.
Five weekdays at one hundred minutes is eight hours and twenty minutes, plus an optional weekend mock. That beats twenty-five hours of tab-open time by a distance, and it is the difference between arriving at interview week sharp and arriving at it hollowed out.
The zero day is mandatory, not a reward
One full day a week with no problems, no editorials, no reading about problems. Fix the day in advance, the same weekday every week, so it never has to be decided while you are tired. Two reasons it is not optional. The first is consolidation: a family you drilled on Thursday is more available on Saturday if Friday was empty. The second is structural. Life will interrupt this schedule, and a plan with slack absorbs the interruption, while a plan with none shatters at the first missed evening. The people who quit in week five almost always spent weeks one to four proving they did not need a rest day.
Four signals you are already past the line
These show up before the collapse does, usually a week or two ahead, and every one of them is checkable tonight.
- You read the solution before you have really attempted. Concretely: the editorial or the discussion tab is open inside the first five minutes and you tell yourself it is only for a hint. The rule that keeps this honest is one hint at minute 25, the full solution at minute 45, never earlier. If you cannot hold that line for three days running, your load is too high.
- The solved count has become the metric. If the sentence in your head at the end of a session is "I did four today" rather than "I can do sliding window from cold now", the counter is running your schedule. Numbers make a fine dashboard and a terrible goal, because the fastest way to raise a count is to stop attempting things properly.
- You feel dread at opening the tab. Not boredom, dread: the small flinch before you click the bookmark, the sudden urge to reorganise your notes instead. It is the earliest signal on this list and the most accurate, and it deserves to be taken seriously the first time you notice it rather than the fifth.
- You have started dropping mocks. A mock is the highest-value and most uncomfortable hour in any prep plan, so it is what a depleted person cancels first. The excuse is always that you are not ready yet. Cancelling one you had already scheduled is the clearest tell on this page.
One signal is noise. Two in the same week means cut to a single block for a fortnight. Three means stop and run the rebuild below, because pushing on from there costs more weeks than the rebuild does.
The rebuild protocol if you are already burnt out
- Days 1 to 14: 25 minutes a day, easy problems only. Set a visible timer. Two weeks, no exceptions on the difficulty, no matter how insulting the problems feel.
- Re-solve problems you already know. This is the part people resist and it is the point. You are not trying to learn anything for a fortnight. You are reattaching the act of opening the editor to the feeling of finishing something, and known easy problems are the only dependable way to do that.
- Stop at 25 minutes, especially when it is going well. Ending mid-flow feels wasteful and is exactly what rebuilds the willingness to start tomorrow. Do not bank extra minutes for later.
- Nothing else counts during those two weeks. No new mediums, no contests, no editorials, no tracking spreadsheet. Hide the streak counter if the site lets you. Keep the weekly zero day through both weeks.
- Day 15: one unseen medium, four-minute clock. Read it and give yourself four minutes to name the family out loud. If you can start it without flinching, go back to the two-block day. If the flinch is still there, run a second easy-only fortnight and ask honestly whether the deadline you are grinding toward is a real one.
If a genuine onsite is six days out, the rebuild is not available to you. Run one block a day plus two mocks instead, accept that prep is over, and put the spare energy into the parts of the loop that are not typing, which is where the marginal hour pays best when you are empty. Working through the other prep guides is a better use of a burnt-out evening than problem 241.
Where this plan is wrong, and where our own tool is
The schedule assumes prep alongside a life. If you are between jobs with a funded runway and an onsite in three weeks, three blocks with a long midday gap is defensible for a bounded period, as long as the zero day and the mocks survive it. The version that fails is the open-ended one with no end date attached. It also assumes coding rounds specifically. It does nothing for system design or the behavioural half, both of which improve by talking rather than typing, and both of which usually return more per hour than another medium once you are past your first hundred problems.
One honest note, since a vendor publishes this page. OfferPilot AI does not raise your tolerance for hours, and running Code Copilot during block one destroys the block outright, because the whole value of those 25 minutes is what you retrieve unaided. Where it belongs is minute 46, after the attempt is over: Ctrl+Alt+S captures the problem, including across several scrolls, and you get an optimal solution with intuition, a dry run and the Big-O to compare against what you actually wrote. It is Windows 10 and 11 only, with no macOS, Linux, mobile or browser version, so on a Mac the useful part of this page is the schedule and nothing else. Its overlay is excluded from screen capture at the operating-system level, so it stays out of Zoom, Teams, Meet and recordings, and that is a claim about screen sharing alone. It does nothing about a webcam, a lockdown browser, a proctor sitting beside you, or a phone camera pointed at your desk, and where an assessment forbids outside help the honest window for any tool is the weeks beforehand. The free tier is five minutes a day with no card, and the paid plans sit on the pricing page.
So, how many hours
Two blocks. One zero day. Stop mid-problem. Re-solve last week's work before you touch anything new, and watch the four signals instead of the counter. The person who gets the offer is almost never the one who logged the most hours. It is the one still practising in week nine.