381 words · original PDF was not stored
Spaced repetition software spent two decades letting users twist an ease factor until the queue felt right. SuperMemo called that number the A-factor. Anki called it ease. Both are a single float that stretches or shrinks the next interval after you press Good.
FSRS estimates retrievability from a memory model, not from an ease factor you twist by hand.
The model tracks two latent variables per card: stability (how long retrievability takes to fall from 100% to 90%) and difficulty (how hard the item is to learn). When you grade a review, FSRS updates both, then schedules the next due date so that retrievability at that date matches the retention you asked for.
The A-factor in SuperMemo is a per-item multiplier on the next interval. FSRS has no equivalent because difficulty and stability are already the model's explanation of why two cards behave differently.
That does not mean you should never override the scheduler. It means the override should be an honest multiplier on the output interval, not a hidden rewrite of the memory state. Call it an A-factor if you want the SuperMemo muscle memory, but store it beside the card, apply it after FSRS speaks, and never feed the scaled interval back into stability.
Request retention is the other legitimate knob. Raising request_retention from 0.90 to 0.95 is the principled way to see cards more often. It shortens every interval because the model is targeting a higher chance of recall.
Demo mode is where teams usually cheat. The naive approach is to multiply intervals down so cards return in seconds. If you compress wall-clock time without compressing the unit of a day, FSRS concludes you recalled the card far too early and inflates stability. The diagnostics screen then shows nonsense: a card you have known for two minutes reports a stability of months.
The fix is symmetric. Redefine the unit of a day in both directions. ts-fsrs sees a faster world; everything you persist stays in real time. You have not faked the schedule. You have changed the clock.
Three presets cover a product, a demo, and a talk: a real day, a one-second day, and a one-minute day. Say out loud which one is on. Judges notice when the numbers still make sense.