Struct Decider 

pub struct Decider { /* private fields */ }

The engine’s handle on an installed backend: the typed helpers every stage uses, the per-run counters behind DeciderReport, and the per-run tool-cause cache.

Implementations

impl Decider

pub fn new(backend: Box<dyn DecideBackend>) -> Self

pub fn with_pair_cap(self, cap: usize) -> Self

Set the per-sweep, per-run pair cap (default DEFAULT_PAIR_CAP).

pub fn pair_cap(&self) -> usize

pub fn calibrated(&self) -> bool

pub fn describe(&self) -> String

pub fn report(&self) -> DeciderReport

This run’s report so far.

pub fn ask(&self, state: Value, questions: &[Ask]) -> Result<Answered>

Ask questions about state in ONE request and validate the answer against what was asked. Every failure is LOP-E051 and is counted on the run’s report; callers drop their stage’s contribution on Err.

pub fn classify_causes( &self, texts: &[String], ) -> BTreeMap<String, CauseVerdict>

The tool-cause classifier (E3): for each distinct free-text cause, the closed-vocabulary cause and its probability when a CALIBRATED backend’s argmax reaches CAUSE_MIN_P; None otherwise (the caller keeps unknown). Cached by string for the run, so one string is asked at most once; at most pair_cap distinct strings are asked per run. A failed request leaves its strings unknown for the run (cached as None, so a later caller does not retry into the same failure).