Back to blog

How to restart studying after a long break

An abandoned deck is not a backlog to clear. It is a measurement of what you still know, and the first week back should be run as a measurement.

Hello there! If you are reading this, there is probably an app on your phone with a red number on it that you have been avoiding since April.

Here is the short version. Do not open the deck and start grinding the pile down. Spend the first week finding out what you still know, and let that result decide whether you resume the deck, rebuild it, or delete it. The badge tells you how long you were away. It tells you almost nothing about what you forgot, and treating it as a debt you owe is the fastest way to quit again by Thursday.

My name is Daniel. I'm a third year engineering science student at the University of Toronto, specializing in machine intelligence, and I also work on Nebulearn, which is a conflict of interest I would rather state here than at the bottom. It also means I have watched a lot of restarts go badly, including my own MAT292 deck, which sat untouched from April until the week I started writing this.

I have already written about why people quit spaced repetition in the first week and about what the streak counter does to a missed Sunday. This page is the far end of that story. You already quit, months ago rather than three days ago, and the deck is still sitting there.

Your due count is a calendar, not a memory test

The number feels like a verdict because you are reading it as "cards you have forgotten." It is not that. It is arithmetic on dates.

Every scheduler works the same way at this level. A card gets a due date of last-review-plus-interval, and once that date passes it sits in the pile and waits, looking exactly like every other card in there. A card on a nine day interval on April 2 went stale in the middle of April. A card that had earned a four month interval was still inside its window in August. The app stacked both into one list and gave you a single total, because that total is the only thing it can actually compute. It has no idea which ones survived.

Forgetting after a long gap is not uniform, and the pile hides the shape of it. The cards you had seen twice are probably gone. The cards you had seen eight times over a year are often still in there and need a cue, not a lesson. That is the difference between retrieval strength and storage strength, which I wrote about in why you cannot remember your correct cards during the final. A three month break wrecks retrieval strength across the board. It does much less to material you drilled properly in the first place.

So your deck right now is not "480 cards of debt." It is 480 cards of unknown state, and you do not clear an unknown state. You sample it.

Week one is a sample, not a sprint

That sample is the actual first session, and it is thirty cards.

Pull thirty cards at random out of the due pile. Random is the part people skip, and skipping it ruins the result. Most apps hand you the oldest cards first, so the default order is biased toward whatever you learned at the start of the term, usually the foundational material you are least likely to have lost. Shuffle, or pick by hand from different parts of the deck.

Then answer them and sort each one into three outcomes. Be brutal about the middle column, because the temptation after a break is to count recognition as knowledge.

OutcomeWhat it felt likeWhat it means for the deck
IntactAnswer arrived in a second or two, unpromptedThe card is fine. One clean pass and the interval can grow again.
RecoverableBlank, then the answer surfaced when you saw itMemory is there, access is not. Needs a few short reps, not re-teaching.
GoneNo recognition, or you had to read the answer to understand the questionFunctionally a new card. This is re-learning, not reviewing.

Those three buckets are the measurement, and the ratio between them is the plan. Say you sample thirty cards from a 620 card deck with 480 showing due, and you get eleven intact, seven recoverable, twelve gone. Eleven out of thirty is a little over a third, so roughly 170 of those 480 cards need one pass and then go back to sleep for months. Twelve out of thirty is 40 percent, so roughly 190 cards are effectively new material wearing an old card's clothes. That is the real job: not 480 reviews, but around 190 re-learns plus a lot of cheap confirmation. At a cap of 20 re-learns a day, you are looking at a week and a half, and you knew that on day one from thirty cards instead of finding out on day nine by giving up.

The five day restart

Once you have the ratio, the week has a shape. Here is the one I use, for a deck the sample said is worth keeping.

Day one is the sample and nothing else. Thirty cards, then close the app. Before you close it, set your new card limit to zero, whatever your app calls it (in Anki, deck options). The instinct on day one is to add cards from the lectures you missed, and that instinct is how the restart dies on day four, because you end up carrying an old deck and a new deck at once.

Days two and three are the gone bucket. Only those. Rate them honestly as failures, because that rating is what rebuilds the schedule. Next week's intervals are computed from what you do this week, not from the ratings you gave in April, which is the one genuinely good thing about coming back to a scheduler instead of a notebook. Stop at the cap even on a day when you feel fine.

Day four is the recoverable bucket, which moves fast because you are re-establishing access rather than teaching anything. It is also the day the deck starts to feel like yours again, which is worth more than the number going down.

Day five is the audit. Go through the cards that failed twice this week and ask one question about each: would I write this card today, from scratch, for this exam? If not, kill it. This is the leech protocol applied to a restart, and a break is the best time to use it, because you have just been handed a free list of every card that was structurally bad. A card that was pattern matched, or too long, or secretly three questions, fails after a gap even when the underlying knowledge is fine. In Nebulearn you can skip a card mid session and ignore it so it stops coming back, which is exactly the job here: after a break, triage is the bottleneck, not review.

Only after day five do you add anything new. For a short gap of two or three days, the minimum viable study day is the right frame, but a multi month restart needs the measurement first or you are guessing at the size of the problem.

When the measurement says delete the deck and start over

Sometimes the sample says the deck is not worth saving, and I would rather say that plainly than pretend every abandoned deck deserves a rescue.

Delete and rebuild when all three of these are true. The gone bucket dominates, so you are re-learning most of it anyway rather than reviewing. The cards were not really yours, either downloaded or typed in a hurry from someone else's summary. And you still have the source, so the slides or the PDF are still on your laptop. Then rewriting 150 cards is not extra work. It is a second encoding pass on material you half know, and it costs less than dragging 480 corpses through relearning steps while resenting every one of them.

Keep the deck when the sample shows a real intact bucket, when the deck is large and mature and mostly written by you, or when the source is gone and the deck is the only surviving record of a course. Also keep it when an exam is close. Re-encoding is the better long term move and the worse move with nine days on the clock.

One thing pushes the decision toward rebuilding: if you were already planning to change apps, you are restarting the schedule no matter what. Review history does not transfer cleanly between apps. You can move the text of your cards, and I covered those mechanics in how to switch from Anki, but intervals and lapse counts do not survive the trip. If the schedule is getting rebuilt regardless, the case for keeping cards you would not write today gets thin.

Where Anki is the better place to do this

The plan above is app agnostic on purpose, but the tooling is not equally good everywhere, and Anki is genuinely the strongest place to run a big restart.

If your deck is three years of your own cards, or it is AnKing, do not migrate and do not delete anything. Anki gives you Set Due Date and Forget on individual cards, per deck new limits, filtered decks to build exactly the subset you want to relearn, and add-ons for spreading a backlog over weeks. For a 5,000 card medical deck that precision is the difference between a plan and a mess. It is free on desktop and AnkiDroid, and about $25 once on iOS. If your queue math was already broken before the break, why your Anki queue is 400 cards is the thing to fix first, because a restart into a badly configured deck reproduces the original failure. And if all you need is one shared set before a Friday quiz, open Quizlet and skip this page entirely.

What none of those tools will do is tell you which bucket you are in. That part is manual, it takes thirty cards, and it is the only step here nothing can automate.

When the restart collapses in week two

Most restarts fail around day eight, when the novelty of being back is gone and the deck is still not clean.

The mechanism is the one I described in the honeymoon crash: the work spikes right as the reward flattens. The restart version of it is guilt driven new cards. You have a good day four, you feel like you are back, so you turn the new limit up to catch up on the lectures you missed, and by day eight you are carrying re-learns and new material at once and the pile is bigger than when you started. Hold the limit at zero for the full week. Those cards will still be addable next Monday.

The other failure is trying to make the number hit zero every day, which I argued against in stop clearing your due pile daily. During a restart it is worse than usual, because the pile holds cards in three different states and treating them as one queue is the exact mistake.

Be clear eyed about what a streak feature can do for you here. Nebulearn has a streak freeze that absorbs a missed day automatically, and on the free plan you earn one every seven days. That is a real answer to a busy Tuesday and no answer at all to a three month gap, which is fine, because nothing should be. You are not rebuilding a counter. You are rebuilding a queue. If you cannot get yourself to open the app at all, the five card floor in how to study with no motivation is a better place to start than any plan with five numbered days in it.

An abandoned deck is not a punishment waiting for you. It is a fairly precise instrument that has been measuring your memory while you were gone, and week one is when you read the dial and decide: resume, prune, or rebuild. That split is what I built the review side of Nebulearn around. FSRS is on by default, so the schedule after a restart comes from the ratings you give this week rather than a number from last term, and skip and ignore exist so the cards the sample condemned can leave instead of following you around. The free tier holds 1,000 saved questions and 40 weekly AI credits, so rebuilding a dead deck from the slides is not a purchase decision.

spaced repetitionconsistencystudy tips