Struct BatchMetrics 

pub struct BatchMetrics {
    pub average_overall_score: Option<f32>,
    pub average_safety_score: Option<f32>,
    pub average_factuality_score: Option<f32>,
    pub average_alignment_score: Option<f32>,
    pub flagged_responses_count: usize,
    pub critical_issues_count: usize,
    pub responses_analyzed: usize,
    pub performance_summary: Option<PerformanceAnalysisResult>,
    /* private fields */
}

Aggregate view of one LLMDebugger::analyze_batch run.

Every average is Some only if at least one analysed response actually carried the sub-analysis it summarises; a batch in which nothing produced (say) a safety analysis reports average_safety_score: None rather than 0.0. Both update_from_report and finalize used to be empty bodies, so every field of every batch report published its Default.

Fields

average_overall_score: Option<f32>

Mean of every analysed response’s overall_score.

average_safety_score: Option<f32>

Mean safety_score over the responses that carried a safety analysis.

average_factuality_score: Option<f32>

Mean factuality_score over the responses that carried one. Currently always None, because FactualityChecker has no fact-verification backend and therefore never produces a factuality score – see FactualityAnalysisResult::factuality_score.

average_alignment_score: Option<f32>

Mean alignment_score over the responses that carried one. Currently always None for the same class of reason – see AlignmentAnalysisResult::alignment_score.

flagged_responses_count: usize

Number of responses whose safety analysis flagged content or detected a harm category.

critical_issues_count: usize

Number of responses whose safety analysis rated the risk RiskLevel::Critical.

responses_analyzed: usize

Number of responses folded in, set by Self::finalize.

performance_summary: Option<PerformanceAnalysisResult>

Implementations

impl BatchMetrics

pub fn update_from_report(&mut self, report: &LLMAnalysisReport)

Fold one per-response report into the running accumulators.

Only sub-analyses that are actually present contribute; a None sub-analysis (or a None score inside a present one) is skipped rather than counted as a zero.

pub fn finalize(&mut self, batch_size: usize)

Publish the averages computed from everything folded in so far.

batch_size is recorded as Self::responses_analyzed; it is deliberately not used as the divisor – see MeanAccumulator.

Trait Implementations

impl Clone for BatchMetrics

fn clone(&self) -> BatchMetrics

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BatchMetrics

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

Formats the value using the given formatter.

impl Default for BatchMetrics

fn default() -> BatchMetrics

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for BatchMetrics

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for BatchMetrics

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.