Leitner Card Box - 3D Cabinet

A desk with a five-compartment card file: cards climb the boxes when remembered and fall back to box 1 when forgotten.

Switch to 2D

Presets

Playback

Deck and workload

Changing a value restarts the deck.

Box intervals

Scale the 1-2-4-8-16 day intervals.

Learner and material

Memory-model constants.

Camera

Quality

Layers

Model checks

The science

Forgetting and the spacing effect

Each card has a memory stability S (days). Without review, the probability of recalling it after a delay t falls as R(t) = exp(−t/S) (Ebbinghaus' forgetting curve in its simplest exponential form). A review at recall probability R succeeds with probability R; on success the stability grows as S → S(1 + g·e·(1 − R)), where e is the card's ease and g the spacing gain. The factor (1 − R) is the spacing effect: a review when the memory has faded a little builds more stability than a review straight after learning, and a review at R = 1 builds none. A lapse keeps only a fraction of the stability, S → max(0.3, f·S).

The Leitner box

Cram studies every card several times a day in the last days before the exam. Fixed interval reviews every T days. Expanding follows the SuperMemo-2 style 1, 6, then interval × EF days (shortened to 1 day after a lapse). Adaptive schedules each card when its predicted recall falls to the target r*, interval = −S ln r*. Leitner keeps five boxes reviewed every 1, 2, 4, 8 and 16 days: a correct answer moves the card up a box, a wrong one sends it back to box 1. For a constant recall probability p the long-run share of cards in box b follows a Markov chain; the page checks the simulation against that closed form.

Honest simplifications: one study session per day (so an interval is a whole number of days and a very weak card is reviewed daily at best); recall is a single probability per card with no context, interference or sleep effects; ease is lognormal (spread 0.35) and all constants (initial stability 6 d, g = 7.5, lapse keeps 35 %, 30 s per review) are illustrative, not fitted to data. Recall 30 days after the exam is pure exponential forgetting with no further study. The numbers show the shape of the trade-off, not a prediction for a real learner.

What to try
  • Press Play and watch new cards arrive in box 1 and climb as you answer correctly; a wrong answer sends a card (red flash) back to box 1.
  • Compare the white bars on the third graph with the box shares: they come from the Markov chain for the recall rate the run itself measured.
  • Choose Perfect recall: every card walks up to box 5 and the daily load vanishes.
  • Try Hard material and Fast boxes: shorter intervals keep recall high but cost more study time.
  • Use the Box close-up camera while cards fly between compartments.
Loading Blender model...
Cards per box over time
Mean recall of the whole deck
Box shares against the Markov-chain prediction
Made with 🟠 Blender · Three.js