Function downgrade_confidence 

pub fn downgrade_confidence(
    prior: &Confidence,
    reason: UnknownReason,
) -> Confidence

Cap prior at the ceiling implied by reason. Never raises confidence — if the prior is already at/under the ceiling it is returned unchanged (only the explanation is appended). Ordering uses ConfidenceLevel’s derived Ord where High < Medium < Low < Opaque (a larger discriminant == less confident), so the capped level is max(prior, ceiling).