Read the explanation
This matrix uses an authored exponential retention model. Passive retention is two raised to minus day divided by half life. For an illustrative two point five day half life, day zero gives one hundred percent, day two point five fifty, and day five twenty five. The bars use four pixels per percentage point, widths four hundred, two hundred and one hundred. These are outputs of an assumed equation, not measured memory or personalized learning evidence. Active intervals reset the elapsed-time origin and multiply half life by a prompt-depth value. At an intervention itself elapsed time is zero, so modeled retention jumps to one. The code has a five percent floor only in its final tail branch. Earlier interval branches return the exponential directly. The default five interventions are days one, three, seven, fourteen and thirty. Starting half life is two point five days and multiplier one point eight. Five multiplications give forty seven point two three nine two days. At day ninety, the final elapsed tail is sixty days. The exponential is about point four one four six, displayed as forty one percent. The illustrative bar uses four hundred pixels per retention unit and ends at about one hundred sixty five point eight five pixels. The hours-saved label is a different calculation: ceiling of ninety divided by two point five gives thirty six assumed passive reviews; subtract five active intervals and multiply by one point eight to give fifty five point eight hours. That conversion factor is authored, not observed time saved. These results do not establish efficacy or expected personal performance. Again, good and easy change the currently selected node’s stability by point eight, one point two or one point five, with a lower floor on again. For an illustrative node starting at one, those single updates produce point eight, one point two and one point five. The bars use two hundred pixels per stability unit, lengths one hundred sixty, two hundred forty and three hundred. The global day-ninety calculation instead reads the schedule, global multiplier and base half life; it does not use these node stability values. Distinguish graph state from the global retention estimate. In the offline environment D three initialization can fail before handlers bind, so written button logic is not proof of working controls. Native outcomes and exports require separate checks. The tool is a simulation and its biological or educational labels remain unverified.