Benchmark

Status: Canonical
Owner: PL-006
Last updated: 2026-07-05

Dataset-driven evaluation of agent behavior — read-only with respect to production runtime.

Principles

Rule	Detail
Isolation	Benchmark code paths do not mutate live orchestrator config
Reproducibility	Dataset version pinned in report
Regression	Every release compares against prior baseline

mermaid

flowchart LR
    DS[Ground truth dataset] --> RUN[Benchmark runner]
    RUN --> RPT[Report JSON]
    RPT --> ACC[Acceptance Lab]

Breadcrumbs: Home → AI → Benchmark