Back to blog

Studying Flashcards Offline: Planes, subways and dead zones

Offline flashcards rarely break because of the cards. They break because the review log has nowhere to go. What to check before you lose signal.

Hello there. If you are staring down a four hour flight, a subway line with no signal, or a cottage with one bar, the question you typed is probably some version of "can I still do my cards."

Short answer: yes, if your app keeps the whole collection on the device. Anki does. Anki's desktop app, AnkiDroid and AnkiMobile store your collection in a file on the machine you are holding, and pushing it to AnkiWeb is a separate step you trigger later. You can board, grade 90 cards, land, and sync from the terminal. Most web-first apps need a connection while you review, and Nebulearn is one of them. I work on Nebulearn, so that is my conflict of interest stated up front, and it is also the sentence where the competitor wins outright: for a stretch of real, no-signal offline time, Anki is the correct app and it is not a close call.

The part almost everybody gets wrong is what they are actually protecting. Nobody loses the cards. Card text is small, and any app that has shown you a deck once has it in memory or in a local cache. What you lose is the review log, and you do not find out until you reconnect. That is why the complaints I see are never "my cards disappeared" and always "I did the session and the queue did not shrink," or "my deck scheduled everything twice." Both of those are logging failures wearing a costume.

Cards are the easy part, the review log is not

A flashcard app is really two things pretending to be one. There is the content, which is a pile of text and images that only ever gets read during a session. And there is the scheduling state, which gets written every time you grade a card.

When you press Good, the app is not just turning to the next card. It is changing that card's due date and interval, updating whatever the scheduler tracks about your memory (under FSRS, that is a stability and a difficulty estimate per card), and appending a timestamped row to your review history. That is a write. The card text is read-only and caches beautifully. The write needs somewhere to land.

A local-first app lands it in a file on your device and reconciles later. A server-first app tries to land it on a server. If the server is unreachable, one of three things happens:

  1. The app tells you immediately and stops. Annoying, honest, no damage.
  2. The app queues the writes locally and flushes them when signal returns. This is the good outcome and it is harder to build than it sounds, because the queued grades have to carry their original timestamps or the scheduler misreads how long you actually held each item.
  3. The app keeps flipping cards because the page is already loaded, sends each grade at a request that never leaves the device, and silently drops them. You finish feeling productive. Nothing happened.

Case three is the one that burns people. I have watched it happen in a browser tab on the Toronto subway: the session was already open before the signal died, so the cards kept flipping for twenty minutes and every grade went nowhere. The tell is that the counter at the top stops moving while you keep answering. If the due count is frozen, stop. You are practicing for free, which is fine, but you are not scheduling anything.

Double grading is the expensive failure, not the lost session

Losing a session sounds like the worst case. It is not. Twenty minutes of retrieval practice with no log is still twenty minutes of retrieval practice, and the memory benefit is real even if the app never found out. You will just see those cards again tomorrow.

The expensive failure is grading the same card twice in a short window. Here is the mechanism. Spacing schedulers infer how well you know something from how long you held it before you got it right. A card answered correctly after eleven days earns a longer next interval than the same card answered correctly after one day, because holding it for eleven days is evidence of a stronger memory.

Now run that unlogged session forward. You graded "renal clearance formula" as Good on the plane and it did not save. You open the app that evening, the card is due again, so you grade it again, five hours after you last saw the answer. You are not testing a memory at that point, you are testing a memory you refreshed this afternoon. Of course it feels easy. You press Good, or worse Easy, and the scheduler hands you an interval you did not earn. That card comes back too late, and you will fail it in three weeks with no idea why. Same trap as pressing Easy on a card you barely recalled, except the app helped you do it.

The correction is small and you have to do it deliberately: when you know a session did not save, re-grade the cards one notch harder than they feel. If it felt Easy, press Good. If it felt Good, press Hard. You are compensating for the fact that the interval you are being offered was computed from a gap that did not really happen.

Two devices and one deck is the version that actually loses data

That BlogTake is the rule, and this is why it exists. Everything above assumed one device. The genuinely destructive case is two.

Sunday night you sync your Anki collection. Monday you fly and clear 90 cards on AnkiDroid with no wifi. Meanwhile you are also on your laptop at your parents' place, with the same collection, and you poke at the same deck for ten minutes because you are bored. Both copies have now moved forward from the same starting point, and neither knows about the other.

Anki is good at this and still cannot fully save you. It syncs incrementally, so two devices that reviewed different cards usually merge fine. The problem is cards reviewed on both. One device's scheduling state wins and the other is discarded, and no screen tells you which. If the copies have drifted far enough that Anki cannot reconcile them, it stops and makes you pick a direction, upload this device's collection or download the server's, and whichever side you do not pick loses its reviews entirely. There is no undo.

So the habit is not "sync often." It is: sync before you leave, pick the device that owns the queue, and do not open the deck anywhere else until that device has synced back. Sync is also not instant, which is the other half of the trap. Airport wifi dropping mid-upload, or closing the app before the spinner finishes, leaves you where you started. Watch it complete.

If the deck is shared rather than your own, the same logic applies with more people in it, which is one reason shared decks go stale.

The pre-flight checklist

Most of what breaks offline is decided in the two minutes before you lose signal, so here is what I actually do while I still have bars.

  1. Open the app on the device you will genuinely use, and let a sync finish. Not the device you meant to use. The one that will be in your hand.
  2. Scroll a few cards in each deck you plan to review. Images, audio and occlusion masks are often fetched lazily, and a card whose image never downloaded is a card you cannot answer.
  3. Name the owning device out loud and then leave the other one alone.
  4. Check your day rollover. Anki lets you set the hour a new day starts, and if you are crossing timezones your app's idea of "today" may not move with you. A long eastbound flight can make one day look skipped and the next look doubled.
  5. Be realistic about volume. If your due pile is 400 cards, you are not clearing it on a phone at a tray table, and pretending otherwise is how you end up rushing grades. A sane per-card pace tells you what actually fits in the flight.

Where the tools stand, since the answer differs by app more than by algorithm:

AppReviews with no connectionWhere the review log lives
Anki desktopYesLocal collection file, AnkiWeb sync is a separate step
AnkiDroidYes, freeOn the phone
AnkiMobile (iOS)Yes, about $25 onceOn the phone
QuizletOffline access has been a paid feature, check the current plan page before you rely on itQuizlet's servers
NebulearnNo offline modeOur servers

The pattern in that table is not about scheduler quality. Anki and Nebulearn can both run FSRS. The split is where the write goes, and that one architectural choice decides your whole offline story. It is also the one thing you cannot fix in settings, which is why "which app handles offline" is a real question. The fuller trade is in Nebulearn vs Anki, including the iOS price that catches people out.

When you cannot review offline, do the other half

Suppose you are on a server-first app and you are about to spend five hours without signal. Reviewing is off the table. That does not make the flight useless, because the two halves of a flashcard workflow have completely different requirements.

Creation needs nothing. Writing questions is pure thinking, and no scheduler is involved. A flight is the best block of uninterrupted time most students get in a month, and turning a lecture into twenty clean questions on paper is exactly the work that never fits into a Tuesday. You type them in when you land, and the deck was built without a single network request.

The other offline-safe move is retrieval with no app at all. Take a blank page and write down everything you can produce about one topic. Blank sheet retrieval is harder than flipping cards because there are no cues, and it tests the thing an exam tests. No log, no sync, no due dates.

A missed day costs less than you think

The reason people try to bank a plane ride of reviews is usually fear, and that fear is the thing this whole page has been circling.

Spacing schedulers are built to absorb late reviews. FSRS takes actual elapsed time as input rather than assuming you showed up on schedule, so a card due Tuesday and answered Friday is handled, not corrupted. One offline day does not wreck a deck. Three is a slightly bigger Monday. The queue getting genuinely out of hand takes weeks of skipping, not a weekend in a dead zone.

What a missed day does break is the thing that has nothing to do with memory: the streak. And a broken streak is a real risk, because for a lot of people it is the moment the habit quietly ends. That is worth solving directly rather than by faking reviews you cannot log, and it is one reason streaks fail as a motivation system when they are all-or-nothing. In Nebulearn a streak freeze absorbs a missed day on its own, nothing to buy and nothing to tap, and the free tier earns one every seven days. So the flight does not cost you the chain. If you are working and studying, the same logic is why a small non-negotiable daily floor survives travel better than a big daily target does.

So the honest summary of "how to study flashcards offline" is this. Your cards were never the fragile part. The review log is, and the damage lands after you reconnect, as a session that vanished or a card scheduled off a memory you had just refreshed. If your study life is mostly tunnels and planes, use Anki, keep the log on your device, and let one device own the queue between syncs. If your study life is mostly normal weeks with the occasional dead zone, the thing worth fixing is not offline access, it is the queue that made you want to hoard reviews: FSRS scheduling on by default so the pile is priced correctly instead of dumped on you, and a streak freeze so the one day you spent at 30,000 feet does not read as a failure. That is the half Nebulearn is built for, and moving off Anki is worth doing with clear eyes about what you give up, because review history does not follow you between apps.

Spaced RepetitionStudy TipsAnkiWorkflow