Enum UtilityEvidence 

pub enum UtilityEvidence {
    Reported,
    Grounded,
}

How much of a success claim has to be demonstrated before it counts.

CAR already draws this line elsewhere — the flagship assistant cross-checks a final summary’s operational claims against same-run tool receipts rather than believing the prose. The same reasoning applies here, and harder: a codebook fitted on runs that merely said they succeeded is a codebook of whatever the model is most confident about.

Variants

Reported

Take OutcomeStatus at face value, whatever backs it.

Grounded

Require at least one evidence item that is not the agent’s own assessment — a tool result, a state change, an external verification, or a product evaluator. A Success backed only by EvidenceKind::SelfAssessment yields None, not 0.0: “nobody checked” is not “it failed”, and recording it either way is a fabricated training label.

Trait Implementations

impl Clone for UtilityEvidence

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for UtilityEvidence

impl Debug for UtilityEvidence

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for UtilityEvidence

impl PartialEq for UtilityEvidence

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for UtilityEvidence