pub struct ScoreWeights {
pub tool_accuracy: f32,
pub loop_avoidance: f32,
pub task_completion: f32,
}Expand description
Weights used to combine per-candidate metrics into one composite score.
Each weight should be non-negative; the service normalizes the weighted
sum by the total weight, so the three values do not need to sum to 1.0.
Fields§
§tool_accuracy: f32Weight applied to the average tool-call match score (AST-style, partial credit) across all tasks in the suite.
loop_avoidance: f32Weight applied to 1 - (loop/stagnation trigger rate).
task_completion: f32Weight applied to the fraction of tasks the agent completed (produced a final answer instead of erroring out).
Implementations§
Source§impl ScoreWeights
impl ScoreWeights
Sourcepub fn composite_of(
&self,
tool_accuracy: f64,
loop_avoidance: Option<f64>,
task_completion: f64,
) -> f64
pub fn composite_of( &self, tool_accuracy: f64, loop_avoidance: Option<f64>, task_completion: f64, ) -> f64
Weighted mean of the three axes, renormalized over those that were
measured. An unmeasured loop-avoidance axis claims no weight rather than
scoring 0.0 or an imputed 1.0.
The single definition of the composite, because the number is computed in two places that must not drift: once per arm from its own runs, and again when two arms are compared over the axes they share. A second copy of this arithmetic is how a comparison starts measuring the scale instead of the pipeline.
Trait Implementations§
Source§impl Clone for ScoreWeights
impl Clone for ScoreWeights
Source§fn clone(&self) -> ScoreWeights
fn clone(&self) -> ScoreWeights
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read more