Supplymentary Materials for "Plausible Is Not Enough: How Reviewers Judge AI-authored Pull Requests"

Abstract -Code review is the checkpoint where AI-authored pull requests (AIPRs) must earn reviewer trust before integration, but it remains unclear whether reviewers apply the same decision logic used for human-authored pull requests (HPRs). We study AIPR review as a problem of trust calibration and evidence ownership through a paired survey of 239 practitioners comparing HPRs and AIPRs across matched review decisions. Results show that perceived AI authorship changes both the level and basis of reviewer judgment. Reviewers enter AIPR review with lower initial trust, with 64.9% rating AIPRs lower than HPRs. They rely less on conventional human-authorship cues such as author track record and subsystem familiarity, and instead weigh human prompter expertise and bounded scope as key trust signals. AIauthorship also raises approval thresholds: reviewers are less willing to rely on tests or explanations alone, more often reason through critical cases, and block approval when uncertainty remains unresolved. Verification effort shifts from trust shortcuts toward defensive validation, including test inspection, contextchecking, local execution, and partial reimplementation. These findings show that AIPR review is not only about generated-code correctness. It is an evidence-ownership problem: reviewers need a defensible chain from generated change to validated behavior and maintainable ownership. Review tools and agentic workflowsshould therefore surface provenance, separate generated claims from human-validated evidence, and make unresolved uncertainty visible, helping reviewers decide whether plausible code is trustworthy enough to merge