Metrics center
Card Health
Card Health checks whether the card structure creates good conditions for a clear review. It does not answer 'do you know this?'; it answers something closer to 'is this card phrased in a way that can test what it intends to test?'.
What Card Health means
Card Health is the structural quality of the review instrument. The evaluation belongs to the current card version and shows whether the prompt, answer, and context create a clear retrieval attempt.
The scheduler does not fix the card
A sophisticated scheduler decides when the card comes back. It does not turn an ambiguous prompt, an oversized answer, or an accidental cue into good review. Timing and card formulation are different problems.
What the policy checks
The current public policy is card-health-v1. It uses deterministic detectors for minimum content, answer shape, cues, clarity, duplicates, and some behavioral patterns when version evidence exists.
Healthy does not mean easy
A structurally healthy card can be difficult and still be forgotten. A problematic card can feel easy because it leaks a cue. Card Health evaluates the question; review evaluates memory.
Warning and critical
Warning means a fragility deserves attention, but it does not block activation by itself in the current contract. Critical means the problem is important enough to prevent release of that version when the queue applies Health.
Health belongs to the version
Editing the prompt or answer changes the evaluated content. The evaluation for version 1 does not prove version 2 is healthy; the new phrasing needs its own canonical reading.
Before activation
Saved cards can remain waiting to enter review. The real flow is created or imported card, current version, canonical Health, activation queue, and release when eligible.
How to repair
The practical path is to understand the issue, revise the phrasing, edit the card, create a new version, re-evaluate, and move toward activation when the contract allows it. Ignoring a warning and fixing a card are different actions.
Before and after
Synthetic examples show publishable problems without using real user content. They teach how to reduce ambiguity, separate ideas, and avoid accidental cues.
What Health does not measure
Card Health does not measure knowledge, retention, global understanding, topic importance, curriculum coverage, approval chance, absolute factual truth, personal difficulty, or the quality of every possible open answer.
Inside Fluorgan
The representation below is synthetic. It shows version status, issue, severity, and repair action without claiming to be a real product screenshot.
Contract and review
The public reference consumes PublicMetricRegistry: policy, version, states, severities, activation implication, dismiss, and last review. Internal paths, SQL, and tables are not exposed.