Risk‑Driven Delivery and Unattended Testing in Baidu Search: Practices and Insights

Risk‑driven delivery is a crucial research direction in Baidu's intelligent testing practice; three observations motivate this work: over 80% of projects have no associated bugs, many test tasks fail to uncover defects, and testers still experience missed issues.

The team will publish three articles to reveal Baidu's risk‑driven delivery; the first focuses on Baidu Search's unattended delivery practice, describing how risk assessment plays a key role in automating the delivery pipeline.

Before introducing unattended delivery, the article reviews the evolution of delivery models from manual testing to semi‑automated and fully automated pipelines, noting that DevOps has become the main delivery mode for search, yet testers still spend considerable effort on failure analysis, environment preparation, and manual hand‑offs.

Three major tester dependencies are identified: decision‑making (which tests to run), process (workflow hand‑offs), and conclusion (risk judgment). Eliminating these requires intelligent execution and risk‑assessment capabilities so that the entire delivery can proceed without human intervention.

Three foundational capabilities are required for unattended delivery: complete testing ability —ensuring test tasks are comprehensive and effective, exemplified by AI‑assisted DIFF testing that combines white‑box data, models, and business logic to automatically evaluate results; stable build ability —governance of underlying dependencies, full‑cycle stability measures, and digitized build metrics to prevent pipeline failures; and precise evaluation ability —a quality‑score model that aggregates features such as code changes, developer profiles, test outcomes, and coverage, producing risk scores that enable automated admission or trigger manual review.

By continuously improving these capabilities, Baidu Search achieves over 90% autonomous testing coverage and more than 40% unattended delivery, freeing testers for higher‑value tasks and significantly reducing per‑requirement delivery costs.

The article concludes with a brief recruitment notice for testing development, Java, C++, mobile development, and machine‑learning engineers at Baidu MEG Quality Efficiency Platform.