pub struct BenchAuditPolicy {
pub min_trust_score: f64,
pub allow_above: f64,
pub margin_slack: f64,
}Expand description
Configuration for BenchAuditPolicy::evaluate.
Defaults are deliberately conservative: a clean suite (bench_trust_score >= 0.85) backed by a delta larger than the flagged-task noise floor will
produce an Allow verdict; otherwise the gate denies or asks for HITL.
Fields§
§min_trust_score: f64Trust below this is an unconditional deny.
allow_above: f64Trust at or above this — and a delta outside the flagged margin — is the only path to an explicit Allow.
margin_slack: f64Multiplicative slack applied to the flagged-task ratio when computing
the noise margin. delta_score <= flagged_task_ratio * margin_slack
denies as “within flagged margin”. 1.0 means use the ratio as-is.
Implementations§
Source§impl BenchAuditPolicy
impl BenchAuditPolicy
Sourcepub fn evaluate(
&self,
claim: &BenchGatedClaim,
summary: &BenchTrustSummary,
) -> Vec<PolicyVerdict>
pub fn evaluate( &self, claim: &BenchGatedClaim, summary: &BenchTrustSummary, ) -> Vec<PolicyVerdict>
Evaluate a single bench-gated claim against the audit summary.
Returns a bag of PolicyVerdicts ready to be fed into
crate::precedence::resolve_conflicts. The caller wires the
resolution into a crate::decision::PolicyDecision with the standard
crate::trace::DecisionTrace — see
crate::engine::PolicyEngine::evaluate_bench_gated_decision.
An empty result vector means “the audit plane has nothing to say” — the caller’s deny-by-default fallback is what then applies.
Trait Implementations§
Source§impl Clone for BenchAuditPolicy
impl Clone for BenchAuditPolicy
Source§fn clone(&self) -> BenchAuditPolicy
fn clone(&self) -> BenchAuditPolicy
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more