# How the measurements work

Quantified Cognition uses your Anki review history to estimate daily recall and
memory formation. It compares your answers with Anki's FSRS predictions, so the
scores account for how likely you were to remember each card and how long it
had been since you studied it.

## Recall: how you performed that day

For each card, the add-on uses your first answer on an Anki day. Again counts
as a failed recall; Hard, Good and Easy count as successful recalls. It compares
each outcome with the predicted chance of remembering and averages the
differences across the day.

A score above zero means you recalled more than predicted; below zero means
less than predicted. Scores are expressed in percentage points. For example,
if 80 successful answers were predicted across 100 eligible cards and you
answered 85 successfully, the score is +5 percentage points.

## Formation: what later reviews tell us about a study day

For each card studied on a day, the add-on looks at your first answer on the
next Anki day when you review that card. It compares that answer with the FSRS
prediction after your last review on the earlier day, allowing for the actual
time between reviews. The difference contributes to the earlier day's
formation score.

If you study a card on Monday and review it again on Friday, Friday's answer
helps estimate Monday's memory formation. The daily score averages the
available follow-ups. It is shown overall and separately for follow-ups within
30 days and beyond 30 days.

Cards awaiting another review are shown as pending, outside the score until
a follow-up is available. Coverage shows how much of a study day's result is
available. Scores become more complete as cards are reviewed again, with
longer follow-ups taking more time to accumulate.

## Sleep stages, naps and activity

Personal insights compare your daily scores with sleep and activity from the
same periods of your history. Recall is compared with the preceding main
sleep. Formation is compared separately with the main sleep before and after
studying. Actual sleep-session times distinguish main sleep from naps, whose
timing, duration and stages are examined separately. Steps use the previous day
for recall and the study day for formation.

Comparisons between light, deep and REM sleep use the same eligible days and
account for total sleep duration. For example, an estimate for ten more minutes
of deep sleep replacing light sleep compares a change in sleep composition
while holding total sleep and REM sleep constant. Oura and Google Health
measurements are analyzed separately.

## Uncertainty and interpreting changes

Scores include review counts, follow-up coverage and uncertainty intervals.
Personal insights also show uncertainty and identify when there is
insufficient data. Estimates depend on both the number of usable reviews and
the number of days they represent.

Existing history is recalculated with your current FSRS settings. This allows
older reviews to be analyzed, including eligible reviews from before FSRS and
reviews synced from other Anki apps. Reviews that cannot be reconstructed
reliably are excluded with a reason. Changing FSRS settings can change past
scores. Dates use your configured Anki day boundary.

These scores estimate learning and recall relative to FSRS predictions. Card
content, study routines, scheduling and conditions at the later review can also
affect them. The comparisons can help identify patterns worth exploring in
personal experiments. Establishing cause and effect also requires an
appropriate experimental design.

The [research information](RESEARCH_INFORMATION.md) explains voluntary
participation, and the [privacy policy](PRIVACY.md) describes the data used and
shared.