Domenic Denicola's essay argues that spaced repetition systems have quietly improved in the last couple of years, and that the biggest change is a better scheduling algorithm. The familiar idea is simple: show flashcards, let the learner grade the response, then decide when to show the card again. The essay's case is that the scheduling layer behind that loop has become much smarter, and that the result is less wasted time and less frustration for people trying to build durable memory over the long term.
The piece starts from an ordinary problem. If a subject is part of your full-time job, repeated exposure comes naturally. If it is a part-time pursuit, or a school subject that gets only a few hours a week, recall gets harder to maintain. Denicola uses his own Japanese study as the concrete example, but he broadens the point to almost any fact-heavy area of life. He even asks whether podcasts and Hacker News stories would feel more useful if the information in them actually stayed with you. That is the niche spaced repetition tries to fill.
The essay then sketches the older way these tools worked. A card might come back after one day, then six days after a correct answer, then 15 days later after another success. The point of that familiar progression is to keep reviews from piling up while still hitting the material before it is forgotten. That model has helped millions of learners, but Denicola says it is no longer the best that software can do. Recent improvements, he argues, make the system less clunky and more adaptive to real memory behavior.
The name he highlights is FSRS, an algorithm by Jarrett Ye. In the essay's telling, FSRS is the main ingredient in the recent jump in quality. Instead of relying on a rigid, one-size-fits-all schedule, the newer approach uses better prediction to decide when a card should return. Denicola describes this as a quiet revolution rather than a flashy product change, because the benefit is mostly invisible: fewer unnecessary reviews, better timing, and a system that feels more responsive to the learner's actual performance.
He makes the practical benefit concrete with a familiar productivity claim: under the right schedule, a learner might add 10 new second-language words a day and still spend only about 20 minutes reviewing. That is not a promise that learning becomes effortless. It is a claim that the software can waste less of your time on cards that do not need attention yet. In a domain where attention is usually the limiting factor, that matters almost as much as raw recall rates.
The broader point of the essay is not just that spaced repetition works. It is that the software stack around it has gotten more capable, and that the best systems now owe a lot to recent work in machine learning. Denicola's post reads less like a product review than a status update on a mature idea that has suddenly become much more useful. For learners who have bounced off older tools, that may be the most important news in the piece.
