CITATION.cff

cff-version: 1.2.0
message: "If you use Plumbline, please cite it as below."
title: "Plumbline: a fail-closed evaluation harness for government-facing chat systems"
abstract: >-
  A deterministic, offline-by-default evaluation harness that grades a
  government-facing chat system against an executable quality bar and produces
  a sealed, provenance-stamped audit report. Fifteen scoring suites, per-suite
  confidence intervals and minimum detectable effect, baseline regression
  comparison, and a pinned fail-closed CI gate. Tampered evidence refuses to
  score rather than scoring around the tampering; a target that returns nothing
  is never credited with a pass; identical inputs and seed produce
  byte-identical reports. The bundled dataset is synthetic and demonstrates the
  instrument; it measures no real system.
type: software
authors:
  - given-names: Chelsea
    family-names: Kelly-Reif
license: Apache-2.0
version: "0.2.0"
date-released: "2026-08-22"
repository-code: "https://github.com/ChelseaKR/plumbline"
url: "https://github.com/ChelseaKR/plumbline"
keywords:
  - evaluation
  - audit
  - llm
  - civic-tech
  - ci-gate