Every question has its own forgetting curve.
For each question, the engine keeps a probability: the chance you still know it. It falls over time. A right answer lifts it back up and lengthens your memory, all the more if you had started to forget.
- Solid: still known at 90% or more a week later, without seeing it again.
- Spacing: a late right answer does more to consolidate than a right answer the next day.
A typo is not forgetting.
A mistake on a question you know well is usually just a letter out of place. The engine knows it: it measures how often you slip and does not throw your memory away over one typo.
- In the log of one learner followed for a month (4,527 answers), half the mistakes on questions already passed were slips of one or two letters.
- Accents, capitals and punctuation: tolerated and flagged by default, required if you choose.
ochtentaexpectedochentaochtentaexpectedochentaWhat you already knew fades differently from what you learn.
A word close to your own language, a rule from school: right the first time, never missed. The engine files these questions apart and measures their memory separately from that of what you learn here.
- Already known: measured module by module, kept close to your average while the module is new.
- A module you already know goes by fast; a module new to you slows down and comes back more often.
Every slot in your session goes to what pays off most in 30 days.
For every question seen and for the next new ones, the engine works out how much answering now raises the chance of knowing it in 30 days. The biggest gains take the twenty slots.
- New and reviews are mixed: new questions come in blocks from one module, slipped between shuffled reviews.
- A question already answered within half a day waits: getting it right twenty minutes later proves little.
The whole level seen in 60 days, then you farm.
Beneath the choice by gain, a floor: enough new questions per session to meet the whole level in 60 days, at your measured pace. Once everything is seen, only reviews remain: you farm until it is anchored.
- The floor is recalculated every session: what is left to see, over the sessions you have left before the deadline.
- No mistakes at all: only new material, at full speed. A module that falters: it goes first.
Six parameters, measured on your answers, all the time.
Nothing to set, no average imposed: everything the engine assumes about you is measured on your answer log, and updated with each answer in a few dozen multiplications.
- Already known · right on first contact, chance on choices removed; per module
- Learning · a question missed, known on another day: what the explanation taught
- Inattention · a mistake on a question you know well
- Immediate memory · right within half a day of a mistake: next to no proof
- Memory of what you learnt · its length and its spacing effect
- Memory of what you knew · the same pair, measured apart
You see what the engine knows about you.
In Progress, the “My engine” panel (advanced interface) shows what it has measured on you and your next session. At the end of each session, the summary shows what your memory of the level has gained.
Almost half the answers for the same result.
Before reaching the app, the engine was put to the test on a bench: simulated learners, with memories, paces and inattention drawn at random, trained for 120 days with the old engine, then with this one.
- The engine sees only what the app sees: right or wrong, and when.
- Even under a forgetting law it does not assume, it still needs 34 to 35% fewer answers.
Let the engine learn your memory.
It starts measuring you from your very first answer; each session fits you a little better.