Crate larc_loops 

larc-loops

Agentic loop convergence traits — strategy-agnostic interfaces for deciding when to halt, continue, or escalate an agentic execution loop.

Provides ConvergenceResult and six convergence strategy traits. Implement whichever fits your loop topology; compose multiple traits for hybrid strategies.

Traits

Trait	Topology
BlastScore	Compound risk/score threshold
ConvergenceGate	Gate-by-gate with phase-back
NPassVerifier	N independent verification rounds
QueueDrain	Bounded queue exhaustion
[InterestDecay]	Simulated annealing cooling
[IntervalWatch]	Interval polling with deadline

Quick start

use larc_loops::{ConvergenceResult, BlastScore};

struct ScoreCheck;
impl BlastScore for ScoreCheck {
    fn check(&self, score: f64, threshold: f64) -> ConvergenceResult {
        if score <= threshold {
            ConvergenceResult::Converged
        } else {
            ConvergenceResult::Blocked {
                reason: format!("blast score {score:.2} > {threshold:.2}"),
            }
        }
    }
}

Enums

ConvergenceResult

Outcome of a single convergence check.

Traits

BlastScore

Convergence via compound risk scoring.

ConvergenceGate

Gate-by-gate convergence with phase-back capability.

NPassVerifier

Convergence via N independent verification rounds.

QueueDrain

Convergence via bounded queue exhaustion.