Adaptive review
Each review creates new evidence about that card.
Fluorgan records the attempt, separates recall from effort, confirms the canonical rating during saving, updates memory, and uses the scheduler to estimate the next review.
Overview
Adaptive review is the sequence real evidence, updated memory, and new estimated interval; it is not AI choosing arbitrarily.
Before reveal
The attempt happens before seeing the answer so retrieval stays separate from recognition or rereading.
Recall before effort
not_recalled closes effort selection; recalled opens high, normal, or low effort.
Canonical rating
not_recalled becomes wrong; recalled with high effort becomes hard; normal becomes ok; low becomes easy. save-review remains the authority before persistence.
Card memory
A confirmed review can update stability, model difficulty, last state, and next review.
Next review
The scheduler uses FSRS as the base for estimating when the card should return; the date can change after each review.
Recovery
A confirmed primary failure can create eligible same-cycle Recovery; it is real, has attempt_role recovery, and does not erase primary.
Review cycle
The current contract preserves primary and, when eligible, at most one same-cycle recovery.
What makes it adaptive
Adaptation comes from the card state changing after a confirmed attempt.
Scheduler and Rhythm
Scheduler asks when this card tends to return; Rhythm asks how the whole workload fits availability.
What it does not do
Safe Ahead, Pretest, and Card Health have their own roles and do not replace adaptive Review.
In product
The interface shows prompt, reveal, recall, effort when applicable, saving, and confirmed.
Limits
It does not read minds, guarantee future recall, find a perfect moment, or remove the need for correct self-reporting.
Next step
After creating cards, understand Recall Integrity, Retention, and Workload without links to unpublished future pages.