Automated Self-Testing as a Quality Gate: Evidence-Driven Release Management for LLM Applications

TL;DR

This paper introduces an automated self-testing framework with evidence-based quality gates for managing the release of LLM applications, ensuring stability and safety through empirical evaluation.

Contribution

It presents a novel, evidence-driven release management system for LLMs that incorporates multiple evaluation dimensions and supports independent validation.

Findings

The framework effectively identified regressions and supported stable quality evolution.

Evidence coverage is the key discriminator for severe regressions.

Runtime scales predictably with suite size, enabling efficient scaling.

Abstract

LLM applications are AI systems whose nondeterministic outputs and evolving model behavior make traditional testing insufficient for release governance. We present an automated self-testing framework that introduces quality gates with evidence-based release decisions (PROMOTE/HOLD/ROLLBACK) across five empirically grounded dimensions: task success rate, research context preservation, P95 latency, safety pass rate, and evidence coverage. We evaluate the framework through a longitudinal case study of an internally deployed multi-agent conversational AI system with specific marketing capabilities in active development, covering 38 evaluation runs across 20+ internal releases. The gate identified two ROLLBACK-grade builds in early runs and supported stable quality evolution over a four-week staging lifecycle while exercising persona-grounded, multi-turn, adversarial, and evidence-required scenarios. Statistical analysis (Mann-Kendall trends, Spearman correlations, bootstrap confidence intervals), gate ablation, and overhead scaling indicate that evidence coverage is the primary severe-regression discriminator and that runtime scales predictably with suite size. A human calibration study (n=60 stratified cases, two independent evaluators, LLM-as-judge cross-validation) reveals complementary multi-modal coverage: LLM-judge disagreements with the system gate (kappa=0.13) are attributable to structural failure modes - latency violations and routing errors - invisible in response text alone, while the judge independently surfaces content quality failures missed by structural checks, consistent with a multi-dimensional gate design. The framework, supplementary pseudocode, and calibration artifacts are provided to support AI-system quality assurance and independent replication.