Four passes per deck. Stems rephrased on every revisit.

Four passes per deck before moving on. Pass 1 is a cold baseline through every card on MCQ format to get a real miss list. Pass 2 is the miss list only, with the stems rephrased so you cannot pattern-match the sentence shape from pass 1. Pass 3 is the still-missed cards switched to free-response format, forcing cold recall instead of recognition. Pass 4 is the last wrong answer's source slide, read only that slide, then back to drilling. Rotate decks, not topics, so each deck reaches a high-confidence state before the next one starts.

Because by attempt three the brain is matching the sentence shape, not retrieving the fact. The card asks 'A 64-year-old man presents with crushing substernal pain radiating to the left arm, most likely diagnosis?' and after three reads you know the answer is C without parsing the stem. On the exam, the same fact arrives as 'A patient describes a heavy pressure in the center of the chest that travels into the shoulder during exertion' and the matching breaks. The retrieval-practice literature calls this the recognition-vs-retrieval gap and it is the failure mode of every static MCQ deck used in a single-night cram.

Hand-authoring is the dead end. Roughly 100 cards per 90-slide deck at one card per minute is an hour or two per deck; you do not have 30 to 60 hours before the exam. ChatGPT will produce questions from a single PDF, fine for one calm afternoon, but it does not enforce a quality rubric on the distractors, does not track which cards you missed, and returns the same wording on the second take. The tool that earns its place in a cram drill auto-generates MCQs against your professor's actual slide deck in roughly 60 seconds per deck and rephrases the stem on every revisit.

Rough budget on a 90-slide deck converted to about 200 MCQs. Pass 1 cold baseline: ~25 minutes (six to seven seconds per card on average, you are guessing fast on the unknowns). Pass 2 miss-only rephrased: ~10 minutes on the 80 to 120 cards you missed. Pass 3 free-response on the stubborn misses: ~8 minutes on the 30 to 50 cards that survived pass 2. Pass 4 source-slide reads on the last 5 to 10 wrong answers: ~5 minutes. About 50 minutes wall-clock per deck. Six decks fits in a six-hour session if you keep moving and stop wandering between decks.

Start MCQ, end on something colder. The reason is that MCQ is recognition with four options visible; the exam might be MCQ but the retrieval skill you build with cold recall transfers down to MCQ, the reverse is weaker. Pass 1 and 2 are MCQ for speed and to surface misses. Pass 3 on the stubborn misses switches to free-response so you have to retrieve the answer with no options on the screen. If the source material has labeled diagrams (anatomy, histology, pathology slides), the image-occlusion format on those specific cards is the highest-yield use of time per card.

Rotate decks, not topics. Each deck completes the 4-pass loop before you start the next. The reason is that cramming dies from boredom and decision fatigue, and rotating within a single deck (mix cardio cards with renal cards from a different deck) raises the cost of every retrieval because you are also switching mental context. One deck at a time, full 4-pass loop, then close it and open the next. The visible end state per deck (tree at full stage) is the loop that keeps the session running past hour three.

Mostly turn them off. Spaced repetition is optimized for spreading the same card across days or weeks. In one night every card is going to be seen multiple times within the same 60 minutes, which is the opposite of what the algorithm is built for. Use mass practice on the miss list during the cram night, then enable the spaced repetition algorithm the next morning for retention of whatever you keep using post-exam.

UWorld is a curated bank against the test blueprint, which is great for board-style exams but does not cover the specific facts your professor put on slide 47 of last week's lecture. Anki is a static deck where the wording does not change, so revisit five in one night is recognition not retrieval. The combination of generating against your professor's actual slide deck, in about a minute per deck, with the stem rephrased on every revisit, is the gap. UWorld covers the boards version, Anki covers the long game, this 4-pass drill covers the actual content your professor will test on tomorrow.

Same 4-pass drill, with two tweaks. First, switch pass 3 from free-response to case-style stems. The case-style format mirrors the test's actual question shape (vignette plus options), which transfers better to test-day performance than abstract free-response. Second, weight pass 1 toward your weakest organ systems first; the recognition-vs-retrieval gap is largest on systems you have not seen since 2nd year, so a baseline on renal or biostats earlier in the night gives you more time to revisit before sunrise.

Single-night cramming retains about 80% one week later with retrieval practice vs about 34% with rereading, per the Karpicke and Roediger paradigm; that gap holds for the 4-pass drill specifically because every pass is retrieval with feedback. The deeper retention curve (one month, three months) requires the same drill spread over weeks rather than one night, which is the spacing effect on top of retrieval. For the exam at 9 a.m., the one-night version is what you have left. For the cumulative final or boards, run the same drill but ship one deck a night across the semester.