Function validate_all_finite_trial_point 

pub fn validate_all_finite_trial_point<I>(
    label: &str,
    values: I,
) -> Result<(), EstimationError>
where
    I: IntoIterator<Item = f64>,

Same check as validate_all_finite_estimation, for a quantity whose non-finiteness is a statement about THIS TRIAL POINT rather than about the configuration.

An inference-only matrix derived from H⁻¹ at the fitted mode – the frequentist covariance, the influence matrix, the weighted Gram – goes non-finite when the curvature at this rho is singular or unstable. That becomes true or false by moving rho, so the outer search should retreat from the point rather than abort, and the difference is carried by the variant instead of being recovered downstream by matching "must be finite" against a list of field names (#2593).

The rendered message is identical to validate_all_finite_estimation’s – TrialPointRefused displays the bare reason – so nothing that reads the text sees a change.