The forgetting-curve calculator
You're paying for lessons and studying new words. The question this page answers: how much of it will still be in your head in 90 days?
Set your numbers. The model does the rest.
A simplified model: fast early decay (~3-day stability) onto a ~15% long-tail floor for unreviewed new vocabulary, following the shape Ebbinghaus measured in 1885 and Murre & Dros replicated in 2015; 90% sustained retention with spaced review. Real curves vary by word and person; the shape doesn't.
Flip the toggle and watch the gap. That gap is the entire argument for spaced repetition; everything else is implementation detail.
What your numbers mean, and the 3-step fix
Leave your email and the interpretation appears right here: why the without-review line flattens so low, what your break-even review dose is, and the 3 cheapest changes to keep what you're paying for. Plus one short email a day for 5 days on the science underneath (one click to stop).
No spam, unsubscribe anytime.
Why the without-review line crawls
The red line does keep inching up, and that's the honest part of the model: even with zero review, a small fraction of words (the memorable ones, the ones you happen to re-meet) survives long-term. Ebbinghaus's own curve flattens near 20% instead of hitting zero.
But look at the slope. Without review you keep roughly 1 word in 6; the rest of every lesson washes out within days. Read that against the money: most of what you pay a tutor to teach you gets bought, lost, and bought again.
Your break-even dose
The green line costs something: reviews. The honest price at your settings is roughly 1 to 2 minutes per new word across its lifetime, front-loaded in the first weeks. At 10 words a day that's a daily review session that starts near 5 minutes and peaks around 15 to 25 before easing off. Cheaper than one extra lesson a month, and it protects all the lessons.
The 3-step fix, cheapest first
- Capture at the source. The 10-minute post-lesson debrief. Words you never write down are words the curve eats first.
- Review by recall on a schedule. Spaced repetition software, graded honestly. Free path: Anki done right. Automated path: the app whose site you're on.
- Start from the frequency core. If you're under ~500 known words, the 650 essentials buy the most conversation per review minute.
Sources for the decay shape and the spacing effect are on the research page. The model above is deliberately simple; FSRS (what the app runs) fits 21 parameters to your actual reviews instead of assuming one curve for everyone.
Verbamor's version of this chart uses your real reviews.
The app's forgetting forecast shows what you'd remember in 30, 60, 90 days if you stopped today, computed from your own history instead of a model of the average person.
Download the app