Forgetting

What Forgetting does

Forgetting decides when to show you a question again. It watches the practice you already do in Live Talk, MCQ, USMLE, SAQ, and Labelling, and schedules each individual question for the day you are most likely to be on the edge of losing it.

It does not write new questions. It re-serves the exact question you answered before, in the agent that created it, so a review is the same task you already know how to do.

Grading happens without a second click

Most spaced-repetition tools ask you to rate your own recall after every card — Again, Hard, Good, Easy. Forgetting does not, because your answer was already graded.

Where the attempt came from	How it becomes a rating
MCQ, USMLE, Labelling	Correct or incorrect
SAQ	The marked score, banded
Live Talk	The score for that objective in the spoken session

Using a hint or revealing the answer caps the rating, so a question you needed help with is not scheduled as though you recalled it cleanly.

The practical consequence: reviews stay scheduled correctly even in a week when you never open Forgetting at all, because every question you answer anywhere in KoiSwarm feeds it.

The scheduler

Scheduling uses FSRS-6 through py-fsrs, the reference implementation maintained by the Open Spaced Repetition project. It is the same algorithm family Anki adopted as its modern scheduler — not an in-house approximation of it.

Every attempt is appended to an event log, and each question’s schedule is recalculated by replaying that log. Two things follow from this:

• The schedule is reproducible. The same history always produces the same due date.
• Scheduler improvements can be applied to your existing history rather than only to questions you answer from that point on.

The default target is a 90% chance of recall at the moment a question comes due.

Choosing what is in scope

Everything you have ever practised is in scope by default. You can narrow it:

• All — every question you have answered.
• Last 30 / 60 / 90 days — a rolling window.
• From a date — everything since the start of the current block.
• Custom range — a fixed start and end.

Use a range when your last exam is finished and its material is no longer worth carrying, or when you want reviews restricted to the block you are currently sitting.

A review session

1. Open the queue. Forgetting shows what is due now, broken down by the agent each question came from.
2. Start a session. Up to five due questions are gathered, preferring questions from the same original set so the session stays on one topic.
3. Answer in the source agent. The session opens in Live Talk, MCQ, USMLE, SAQ, or Labelling — whichever produced those questions.
4. The result is recorded automatically. Your grade updates the schedule, and the next due date is set.

If nothing is due, Forgetting says so rather than inventing filler.

The optional reminder

You can have Forgetting send one Telegram message per day, at a time and timezone you set, telling you what is due. It is capped at once daily, it is clearly labelled, and it is off unless you turn it on.

When a question disappears

Forgetting stores a reference to your original question, never a copy of it. If you delete the set that question belonged to, the scheduled review is marked unavailable and dropped instead of resurfacing content you removed. Reviews are also always checked against your own account, so a review can only ever contain your own material.

Availability

Forgetting is being enabled one source agent at a time rather than all at once, so the set of agents feeding your queue may still be growing. If a recently practised agent is not yet contributing, it has not yet been switched on.

Interpreting the result

A due date is a prediction about your memory, not a statement about the medicine. The underlying question and explanation carry the same caution as when they were first generated: check anything clinically important against an authoritative source.