Free spaced-repetition flashcards for the Claude certification exams. One fact per card, drawn from the concept library, weighted towards the traps rather than the definitions. Get one right and it comes back later; get it wrong and it comes back sooner. This page explains exactly how that decision is made, and shows you real cards, before you make an account.
A card returns when we estimate you are about to forget it. Not before, because that teaches you nothing.
No black box. This is the whole rule, and it is four lines long.
A new card starts at about a day. That is the gap before we expect to need to show it to you again.
One day becomes two, then four, then eight. A wrong answer cuts the interval in half, so a card you keep missing keeps arriving - which is the point.
We schedule the next showing for the moment your estimated chance of remembering has fallen to about half. Reviewing something you still know solidly is wasted effort; reviewing something you have completely lost is re-learning it.
The interval never drops below six hours, so a card you are struggling with cannot spam you all afternoon, and never exceeds a year.
One consequence worth knowing, because it changes how you should answer: because right doubles and wrong halves, a wrong answer costs you two steps and a right answer buys you one. Guessing to keep a streak alive is expensive. If you do not know it, say so - that is the answer the scheduler can actually use.
Front, then back. This is the shape of every card - one fact, and the reason it matters, not a definition you could have guessed.
A stop condition tied to progress, not just to iteration count. Capping turns bounds the damage; it does not detect that the agent has stopped getting anywhere.
Read the concept behind it →Tool descriptions all compete for the same attention. Past roughly two dozen, selection accuracy degrades - the model is choosing from a menu it can no longer read carefully.
Read the concept behind it →A prefix of the request, not the response. Any change to that prefix - even a timestamp - invalidates the cache from that point on, so ordering the stable content first is the whole technique.
Read the concept behind it →Cards test recall. They do not teach a thing you have never met, and using them that way is how people end up memorising sentences they cannot apply. The order that works: learn it on a concept page, see it work in a production scenario, then let the cards keep it alive while you work on something else. When your mock score is close to the 720 bar, cards are what stop the early material decaying while you chase the late material.
Not sure where you stand yet? The 10-minute diagnostic needs no account and names your weakest exam domain, which is the honest place to point the cards.
The flashcard tool needs a free account - it has to remember your answers to schedule them. Everything on this page, and every concept page it links to, is open.
Every Q is phrased as a real Google search query. Answers cite the same evidence-tagged sources used elsewhere on the site.