The schedule isn't one size fits all
Open a flashcard app that reviews vocabulary on a fixed schedule, day one, day three, day seven, day fourteen, and every word you've ever added gets pushed through the same calendar regardless of whether you've known it for months or met it twice last week. That assumes forgetting happens at the same rate for your tenth repetition of “casa” as it does for the one time you saw “desafortunadamente.” It doesn't. A word you've already recalled correctly five times in a row is nowhere near as fragile as one you failed yesterday, and reviewing both on the same clock wastes time on the word you already hold while arriving too late for the one that's about to slip.
What the algorithm is actually tracking
FSRS, short for Free Spaced Repetition Scheduler, is the open-source algorithm behind the review timing in apps like Anki and, inside HabloHack, behind the flashcards built from words you tap while reading. Instead of one fixed interval, it keeps three separate numbers for every word you've learned, and recalculates them after each review.
Retrievability: the odds you'd get it right today
Retrievability is the easiest of the three to picture. Ask yourself: if you were tested on this word right now, what's the probability you'd recall it correctly, and you get a number between zero and one. Anki's own documentation of the algorithm defines it plainly, as “the probability that the person can successfully recall a particular piece of information at a given moment.” That number is highest right after a review and drops continuously as days pass without seeing the word again, which is exactly why a fixed calendar is a poor fit. The speed of that drop is different for every word.
Stability: how long that probability holds up
Stability is what turns retrievability into something a scheduler can actually use. It's defined as the number of days it takes retrievability to fall from 100 percent down to 90 percent after a review. A word you just learned might have a stability of three or four days, meaning you're already at meaningful risk of forgetting it within a week. A word you've correctly recalled five separate times might have a stability measured in months. Say a word sits at a stability of ten days: on day nine, retrievability might still read around 92 percent, comfortably above a typical 90 percent target, so the review stays queued for day ten or later. That number, not “it's been a week,” is what decides when a given word comes back.
Difficulty: why two words with the same history behave differently
The third number is the reason a cognate like “información” and an irregular form like “anduvieron” don't gain stability at the same rate even after the same number of correct reviews. Difficulty tracks how much a correct recall is allowed to raise a word's stability, and it stays within a fixed range for the life of the word. A failed review pushes difficulty up. An easy, confident recall nudges it back down. Harder words need more successful reviews to earn the same stretch in their next interval.
One knob you control: desired retention
All three numbers feed into one setting a learner can adjust: the desired retention rate, the minimum probability of recall the system tries to keep every word above. Push that target up toward 95 percent and the schedule reviews more often, holding more words closer to perfect recall at the cost of time. Drop it toward 85 percent and intervals stretch further, at the cost of occasionally hitting a word just after it's already faded. There's no single correct setting here, only a trade-off between time spent reviewing and how solid each word stays.
Where FSRS came from
Older scheduling systems, the ease-factor model used by early versions of SuperMemo and still the default in plenty of flashcard software, work off a single adjustable number per card. Get it right, multiply the interval by that factor. Get it wrong, drop the factor by a fixed amount, typically 20 percentage points, and start over. Every card's entire history gets reduced to one number. FSRS replaces that with the three-variable model above, trained on millions of logged review outcomes from real flashcard sessions rather than a fixed formula, and maintained as an open-source project that Anki has adopted as its own default scheduler. Anki's own documentation notes that internal testing found it performing roughly on par with SM-17, the proprietary scheduler behind SuperMemo, one of the original spaced repetition programs, built in the late 1980s.
From how many times to when you'll forget
The practical difference shows up in what actually triggers a review. A fixed-schedule app asks whether seven days have passed. FSRS asks whether this specific word's retrievability has dropped below the retention rate you're aiming for. Those two questions produce different review queues almost immediately. A word you nailed on your first four attempts gets pushed further out each time, because its stability keeps climbing and there's no reason to interrupt you over something you already have. A word you keep fumbling gets pulled back sooner and more often, because its stability stays low. None of that depends on how long the word has been on your list, only on how reliably you, specifically, have recalled it.
What a failed review actually does
Failing a review doesn't send a word back to square one. The new stability after a lapse gets calculated from the word's difficulty and its stability right before the failure, so a word that had built up months of stability before you missed it comes back faster than one you've never gotten right at all. The system isn't just logging a wrong answer, it's folding that failure into everything it already knows about how that specific word behaves for you.
What this looks like across a full set of words, not one card
Multiply that across two hundred words and the gap between the two approaches gets harder to ignore. A fixed calendar keeps dragging you back through words you solved weeks ago at the same rate it reviews the ones actively slipping, because it keeps no separate record of which is which. A per-word model spends that same ten minutes almost entirely on the words at risk, while the words you've locked in surface rarely enough that the review queue doesn't keep growing just because your vocabulary does.
Reading inside that schedule
None of this is an argument for drilling flashcards in isolation. A word's stability still climbs faster when you meet it inside a sentence you're actually reading than when you see it alone on a card. That's the part HabloHack is built around: stories graded by CEFR level from A1 through C2, so you can tap any unfamiliar word mid-page to see its meaning, pronunciation, and an example without losing your place, then add it to a personal vocabulary list, an official one, or one built by other learners. From there, those words become flashcards scheduled with FSRS, tracking each word's own stability and difficulty instead of running every word you've ever added through the same fixed calendar.