Trust Center

A card is grounded when its answer is supported by the specific source segment it cites. We run a deterministic evidence check that compares the card text against the cited segment using token overlap.

Cards fall into one of three trust states:

• Verified — the cited evidence directly supports the answer.
• Needs review — evidence is missing, low-confidence, or inferred. The card is still usable but should be checked.
• Rejected — the answer contradicts the cited evidence or fails moderation.

Model confidence is a prioritization signal, not a trust signal. A card can have high model confidence but fail evidence support, in which case it is marked needs review.

Every generated card records its source document, the specific segment it was derived from, and the generation run that produced it. When you edit a card, the original generation record is preserved. Splits and merges inherit evidence from parent cards.

If a source changes (for example, a URL is updated), affected cards are marked stale and you are offered a review flow rather than silently leaving outdated evidence in place.

If a card has an unsupported answer, incomplete evidence, is a duplicate, covers the wrong concept, or has formatting issues, you can report it. Reports are categorized and routed to the card owner and (for public hub decks) to the hub review owner.

Report categories: unsupported answer, incomplete evidence, wrong concept, duplicate, unsafe content, formatting issue.

Mneva generates original study aids from source material. Cards are not reproductions of exam questions or source text. Each public hub lists its sources and licenses. Approved-source ingestion is restricted to sources with compatible licenses.

Private source content never appears on public surfaces. Public hub pages show source metadata and evidence summaries, never private excerpts or notes.

Public hubs are re-reviewed on a schedule based on report volume and source changes. Stale or reported content is visibly downgraded or unpublished until reviewed.

Public hub decks display a quality score with a visible breakdown: evidence coverage, learner outcomes, reports, freshness, and verified author status. Scores are not based solely on download counts. Learner outcomes require a minimum cohort size before they are shown, to protect privacy.