Dashboards measure how you review. They cannot measure what you are reviewing.

Question quality is a property of each individual card: is the stem correctly stated, are the distractors plausible and parallel, does the answer key match a real authority, does the card type match what the underlying material can support. A flashcard dashboard is a property of the review LOG: how many cards are due today, what your seven-day retention rate is, how many cards are mature versus young, what your weekly streak looks like. The two layers do not see each other. A dashboard cannot tell you that the card you just answered correctly is testing a fact your professor did not teach, and a card-quality eval cannot tell you whether you actually reviewed the deck.

Retention prediction in those tools is conditional on the cards. The published correlations between Anki mature-card count and Step 1 score (the AnKing study, Bonner et al.) hold because AnKing is a community-curated, fifteen-year-old deck where the card quality work has already been done by thousands of edits. The dashboard is reading a clean signal. The same dashboard on a deck of self-generated AI cards with 23.5 points of variance in distractor quality is reading mostly noise, because the underlying cards are not uniform on the quality axis. Retention metrics presume card quality, they do not measure it.

No. Anki, Brainscape, Quizlet, Cram, Tinycards (sunset), StudyBlue (sunset), Mochi, RemNote, and SuperMemo all publish review-side metrics: retention, due cards, mature counts, ease graphs, heatmaps, streaks. None of them publish a held-out evaluation of card content against an item-writing rubric. The only consumer tool with a published per-tool card-quality leaderboard is Studyly's eval at studyly.io/quality, scored on factual correctness, clarity, distractor quality, and question-type coverage on a held-out three-document set.

A student uploads a 90-slide cardiology deck into a generic AI quiz generator that does not run a quality gate. The tool produces 200 multiple-choice questions. About a quarter of those have textbook drift (a fact from the model's training that contradicts the slide), about a fifth have length tells (the correct option is visibly longer than the distractors), and about half have at least one filler distractor (a near-duplicate of the correct answer). The student imports the deck into Anki. After three weeks of reviews the Anki dashboard reports 91% retention and a mature-card count of 178. The dashboard says the student is in great shape. On the class exam the student misses every question whose right answer was on a card with textbook drift. The dashboard never had access to the signal that mattered.

Because the engineering budget at a small team has to land on one of the two layers, and dashboards are deeply solved while card quality is not. Anki and SuperMemo have shipped retention forecasting and scheduling for two decades. Replicating that with a heavier UI would not change any student's exam score. The card-quality layer is where the room for improvement still exists: the spread between the best and worst AI generator on the held-out eval is 23.5 points (Studyly 81.3 to Turbolearn 57.8), and the mechanism that decides the score (source-grounded distractor pool versus pretrained free-association) is replicable. So the bet is to spend the budget there and keep the dashboard to one tree per deck, which is enough to drive daily-review behavior without competing with the card layer for attention.

Four. Factual correctness: does the answer key match the source document the card was generated from. Clarity: is the stem unambiguous, no double negatives, no compound questions hiding two facts under one stem. Distractor quality: are the wrong options plausible, parallel in grammar and length, free of filler templates ('all of the above', 'none of the above'), drawn from a real distractor pool rather than pretrained free-association. Question-type coverage: does the output mix recall, comprehension, and application question types in proportions appropriate for the source. The rubric is the same one the NBME Item-Writing Guide describes, applied to the AI output as if it were a human-written item bank.

Yes, but the realistic configuration is two tools, not one. Use a source-grounded AI generator that publishes a card-quality eval for the card layer (your class lectures, where no community deck exists). Use Anki itself for the dashboard layer (mature-cards count and retention forecasting on the cards you import). Studyly exports to .apkg with a non-colliding note type so the cards land in your Anki collection alongside AnKing without breaking either. Anki's dashboard then reads a clean signal because the imported cards have already cleared the quality rubric at generation time.

It is a difficulty signal, not a quality signal. Confidence-based repetition (CBR) asks you to rate how confident you were in your answer on a 1 to 5 scale, then schedules reviews based on the rating. The score the dashboard surfaces is a function of YOUR ratings, not of whether the card itself is well-written. A card with a length-tell distractor will get high confidence ratings because the test-wise student picks the right option without knowing the material; the CBR score will move the card to a long-interval queue; the dashboard will report mastery. The card is still bad, the dashboard still reads it as mastered, and the exam still tests the underlying fact.

No. Quizlet's UI runs on user-generated cards, and the variance there is wider than on any AI generator because the source population is millions of high school and undergrad students. The relevant comparison for graduate study is between AI generators with different quality gates (or no gate), and the leaderboard at studyly.io/quality is what that comparison looks like under a held-out rubric. Quizlet is a different category. The category the keyword is asking about is the dashboard-quality layer of any flashcard tool, not Quizlet specifically.

Methodology, the four criteria, the three held-out documents, the per-criterion breakdowns, and the per-tool scores are at studyly.io/quality. A worked example of why source-grounded distractor pools land where they do on the eval, plus a 90-second-per-card hand-grading rubric you can run on any deck yourself, is at studyly.io/t/anki-card-distractor-quality.