https://koiswarm.com/blog Experiments on memory, review timing, question formats and exam preparation, written for medical students. en Sat, 12 Sep 2026 00:00:00 GMT https://koiswarm.com/blog/does-teaching-help-you-learn https://koiswarm.com/blog/does-teaching-help-you-learn Sat, 12 Sep 2026 00:00:00 GMT Speaking The advice is everywhere and the research is real, but it is narrower than the slogan. What actually produces the effect is not teaching — it is being made to explain without your notes, and then being asked why. https://koiswarm.com/blog/no-one-to-study-with https://koiswarm.com/blog/no-one-to-study-with Sat, 12 Sep 2026 00:00:00 GMT Study strategy A study partner gives you four separate things, and only one of them requires another human being. Knowing which is which tells you what you are actually missing, and what you can replace. https://koiswarm.com/blog/studying-out-loud-vs-in-your-head https://koiswarm.com/blog/studying-out-loud-vs-in-your-head Sat, 12 Sep 2026 00:00:00 GMT Speaking Saying material aloud does beat reading it silently, and the effect has a well-documented catch: it mostly works on the items you actually spoke. What that means for how you use it. https://koiswarm.com/blog/talking-reviews-vs-clicking-cards https://koiswarm.com/blog/talking-reviews-vs-clicking-cards Sat, 12 Sep 2026 00:00:00 GMT Comparison We costed a day's review queue both ways. Speaking is not faster — for the same measurement precision it takes about a quarter longer. It is still the better trade, for a reason that has nothing to do with speed. https://koiswarm.com/blog/ai-tools-international-medical-students https://koiswarm.com/blog/ai-tools-international-medical-students Fri, 11 Sep 2026 00:00:00 GMT Study strategy If you are studying medicine in English but did not grow up in it, most study advice was written for someone else's situation. Here is what each tool actually solves, what it costs, and what to use in which year. https://koiswarm.com/blog/surviving-lectures-in-a-second-language https://koiswarm.com/blog/surviving-lectures-in-a-second-language Thu, 10 Sep 2026 00:00:00 GMT Study strategy A practical playbook for the ninety minutes themselves: what to do the night before, what to do while you are losing the thread, and what to do the same evening. Mostly free, mostly not about software. https://koiswarm.com/blog/studying-medicine-in-a-second-language https://koiswarm.com/blog/studying-medicine-in-a-second-language Wed, 09 Sep 2026 00:00:00 GMT Study strategy Translation apps fix words. They do not fix the thing that is actually costing you marks. What the language barrier really does to studying, and which kind of tool addresses it. https://koiswarm.com/blog/forgetting-agent-release https://koiswarm.com/blog/forgetting-agent-release Tue, 08 Sep 2026 00:00:00 GMT Release Reviews now schedule themselves from the practice you already do. No card writing, no rating your own recall, no separate deck to maintain — and FSRS-6 doing the scheduling underneath. https://koiswarm.com/blog/how-koiswarm-compares https://koiswarm.com/blog/how-koiswarm-compares Fri, 04 Sep 2026 00:00:00 GMT Comparison A straight answer for medical students. What each tool is best at, what it really costs, and which one to use depending on whether you are revising your own lectures or three months out from Step 1. https://koiswarm.com/blog/sm2-versus-fsrs https://koiswarm.com/blog/sm2-versus-fsrs Tue, 01 Sep 2026 00:00:00 GMT Comparison SM-2 is the algorithm behind most spaced repetition medical students have ever done. Given the same cards and the same memory, it needed 21% more reviews — and a quarter of them landed on cards that had barely been forgotten. https://koiswarm.com/blog/the-two-percent-problem https://koiswarm.com/blog/the-two-percent-problem Fri, 28 Aug 2026 00:00:00 GMT Sources Per-claim accuracy is the wrong number to look at. We worked out what happens when small error rates meet large note sets, and what it costs to check the result with and without a page reference attached. https://koiswarm.com/blog/what-speaking-reveals https://koiswarm.com/blog/what-speaking-reveals Tue, 25 Aug 2026 00:00:00 GMT Assessment We measured the difference between right-wrong marking and a graded rubric using the model behind partial credit. Score levels help — but they stop helping after four. What actually doubles the signal is marking two dimensions at once. https://koiswarm.com/blog/dead-distractors https://koiswarm.com/blog/dead-distractors Fri, 21 Aug 2026 00:00:00 GMT Assessment A five-option question where two options are obviously wrong is not a five-option question. We measured what implausible distractors cost: a quarter of the item's information, and twenty-two extra questions to learn the same amount about yourself. https://koiswarm.com/blog/what-a-question-format-measures https://koiswarm.com/blog/what-a-question-format-measures Tue, 18 Aug 2026 00:00:00 GMT Assessment We scored four question formats with the same psychometric model real licensing exams use. A free-response item carries 1.5× the information of a five-option MCQ — and a student who genuinely knows half the material scores 60% on one and 75% on the other. https://koiswarm.com/blog/what-two-thousand-concepts-cost https://koiswarm.com/blog/what-two-thousand-concepts-cost Thu, 13 Aug 2026 00:00:00 GMT Workload The usual objection to spaced repetition is that the reviews pile up until they bury you. We simulated collections from 200 to 2,000 concepts and measured the steady-state daily load. It is eleven minutes. https://koiswarm.com/blog/reviewing-too-early https://koiswarm.com/blog/reviewing-too-early Sun, 09 Aug 2026 00:00:00 GMT Memory The same card, the same effort, reviewed at two different moments. One buys you four days of durability; the other buys twenty. We measured how much memory a single review actually purchases as a function of when you do it. https://koiswarm.com/blog/when-to-start https://koiswarm.com/blog/when-to-start Wed, 05 Aug 2026 00:00:00 GMT Planning Retention is not what changes when you start revising late — workload is. We measured the peak single-day load for the same 800 concepts across eight different start dates, and it ranges from nine minutes to two hours. https://koiswarm.com/blog/cramming-wins-the-exam https://koiswarm.com/blog/cramming-wins-the-exam Fri, 31 Jul 2026 00:00:00 GMT Study strategy We simulated 800 concepts through a twelve-week run-up, twenty-five times each, comparing a final-week cram against distributed review. Cramming scored higher on exam day — and cost 66% more work to build memories that decayed three times faster. https://koiswarm.com/blog/what-you-forget-before-the-exam https://koiswarm.com/blog/what-you-forget-before-the-exam Fri, 24 Jul 2026 00:00:00 GMT Memory We modelled the decay of a single well-understood lecture. Thirty days later, just over half of it is still reachable — and one review on day one changes that number more than anything else you can do.