Metrics center
Recall Integrity
Recall Integrity separates the result of remembering from the effort needed to reach the answer. First it matters whether retrieval happened. Only then does it make sense to ask how much mental work it required.
What Recall Integrity means
The central question is: what actually happened when you tried to remember? Fluorgan records whether the answer appeared before you saw the back, and, when it appeared, the declared level of effort.
Why not just Easy, Medium, and Hard
A single scale can mix retrieval failure with retrieval effort. 'It was hard' can mean you did not remember at all, or that you remembered correctly after considerable effort.
Recall first, effort second
The public path is simple: prompt, mental attempt, revealed answer, outcome, and effort only when recall happened. Effort for not_recalled is not a UI detail; it has no semantic meaning.
Outcome versus effort
Outcome answers retrieval success or failure. Effort refines a retrieval that happened. Effort is not confidence, and outcome is not an objective semantic accuracy evaluation.
How it reaches canonical rating
The current contract transforms outcome and effort into the canonical rating used by review: not_recalled becomes wrong; recalled with high, normal, or low effort becomes hard, ok, or easy.
Recovery is a real attempt
When a confirmed primary attempt fails, the flow can offer a short recovery attempt in the same session when the scheduler allows it. That attempt can update memory through the normal flow; it is not animation or cosmetic repair.
The failure does not disappear
The primary attempt and Recovery stay attached to the same cycle. If the first attempt was not_recalled and Recovery was recalled, history does not become 'I always knew it'.
Why observed retention does not inflate
Recovery is cognitively real, but it should not inflate first-pass retention. In the public example, 5 primaries with 4 recalls and 1 failure still produce 4/5, even if Recovery from that failure is recalled.
Other evidence has its own contract
Pretest, Contrast Probe, and Variants are not the same as a normal primary review. Each has its own origin, memory, and eligibility, so it should not be treated as the same evidence.
What the system cannot know
Recall Integrity does not read your mind, detect lying, know whether outside help appeared, or automatically verify that every mental answer was semantically perfect.
Inside Fluorgan
The synthetic representation shows prompt, reveal, outcome, effort when applicable, and saving. The attempt counts for memory and progress only after saving is confirmed.
Contract and review
The public technical reference consumes PublicMetricRegistry: version, outcomes, efforts, mapping, cycle, primary/recovery, Recovery, relation to observed retention, and last review.