Struct BatchRecord 

pub struct BatchRecord {
    signature: String,
    rescued: bool,
}

Names the batch LoopDetector::check just counted.

LoopDetector::record_results takes one so it cannot be handed the wrong batch. The answers it records belong to the batch that just ran, and the comparison at the next check is against those; passing the previous batch’s instead would invert the measurement silently, which is a mistake this codebase has made once already with a global id map and caught only in review. Consumed by record_results and deliberately not Clone: recording one batch twice would let a second, different answer rescue a run that never changed its answer, and no call site has a reason to do it. The type is what stops that rather than a rule in prose.

Fields

signature: String

rescued: bool

Implementations

impl BatchRecord

pub const fn rescued(&self) -> bool

Whether this occurrence survived only because an earlier answer moved.

True when the run has repeated more times than the threshold allows and LoopDetector::check passed anyway — which is exactly the turn that signature-only counting would have refused. Known when the verdict is reached, not when the next answers arrive: the reset that saved this turn happened on a previous one.

Trait Implementations

impl Debug for BatchRecord

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

Formats the value using the given formatter.