Read the explanation
This simulator computes an assumed retention curve, not measured personal learning. Initial stability is one divided by decay rate. Unit one uses point zero eight decay and point eight difficulty, so stability is twelve point five days and the effective exponential time constant is fifteen point six two five. Before a review, retention equals initial recall times the exponential of minus elapsed day over that time constant, clamped between five and one hundred percent. An illustrative initial recall point nine five becomes about point three four nine five at fifteen point six two five days with no review. Bars share four hundred pixels per retention unit, widths three hundred eighty and about one hundred thirty nine point eight. This time constant produces exponential decay to one over e; it is not the half-life. The adaptive schedule computes an interval as minus log of the target floor times stability divided by difficulty. It rounds to a whole day and enforces at least one day. With target point eight and the same unit one parameters, the first interval is about three point four eight six six days, rounded to three. Its schedule multiplier is one plus two point five times one minus target, plus point eight: two point three. The next computed intervals round to eight and eighteen, so review days accumulate to three, eleven and twenty nine. The next review is beyond day thirty and omitted. The bars show intervals three, eight and eighteen at twenty pixels per day: sixty, one hundred sixty and three hundred sixty. Whole-day rounding means the nominal target is not an exact enforced floor in the curve. Curve simulation updates stability from actual exponential recall at each review, whereas schedule generation uses the nominal target. These are separate calculations. For an illustrative pre-review recall point eight, the curve multiplier is two point three; at point six it is two point eight. Bars use one hundred twenty pixels per multiplier unit, widths two hundred seventy six and three hundred thirty six. The curve restores recall to one after each processed review. Its efficiency label divides ten times final recall by the larger of one and point eight times session count, capped at ten. That is an authored score, not measured efficiency. JSON includes unit parameters, review days and the sampled curve. CSV writes the curve with percentage recall strings. Offline p5 availability determines whether the draggable chart can render; native numerical behavior needs separate evidence.