pub struct OutcomeInput {Show 13 fields
pub rec_hash: String,
pub target_ref: String,
pub metric: String,
pub baseline: f64,
pub current: f64,
pub unit: String,
pub higher_is_better: bool,
pub baseline_kind: String,
pub baseline_run_id: Option<String>,
pub best_before: Option<f64>,
pub tolerance: f64,
pub current_run_id: Option<String>,
pub cost: Option<CostRead>,
}Expand description
One applied recommendation due for outcome review, with its metric already
re-measured by the engine (which owns the &mut substrate). The outcome
analyzer makes the deterministic changed/regressed decision over this — I/O
in the engine, judgment in the analyzer.
Fields§
§rec_hash: String§target_ref: String§metric: String§baseline: f64§current: f64§unit: String§higher_is_better: boolCarried from the metric snapshot so the analyzer applies the SAME
direction the engine did — see recommendation::is_regression.
baseline_kind: StringWhere baseline came from (OutcomeResult::baseline_kind).
baseline_run_id: Option<String>The evalset run baseline was read from, when it was read from one.
best_before: Option<f64>The best value before the apply (OutcomeResult::best_before).
tolerance: f64The minimum effect size the engine judged under, in the metric’s unit — passed through so the revert draft applies the SAME floor.
current_run_id: Option<String>The evalset run current was read from, when it was read from one.
cost: Option<CostRead>The cost bound’s reading, when the policy set one
(OutcomeResult::cost). A breached bound with quality held is the
advisory-Flag case; with quality regressed it rides on the revert.