SonarQube audit finds Lovable, v0, and Replit have distinct, non-interchangeable quality failure modes.

Nine web applications, three per tool from a single identical generation prompt, run through SonarQube for issue counts, severity distribution, remediation effort, cyclomatic and cognitive complexity, and duplication (arXiv 2608.16302). Lovable concentrates lower-severity issues but carries substantially higher code-smell density per KLOC; v0 and Replit produce more code with more aggressive severity profiles. Small sample and preliminary, and the authors say so. Still one of the few controlled same-prompt comparisons published, and it argues tool choice is a structural trade-off rather than a productivity preference.