Our summary
JevTest separates exploration from proof. A tester supplies controlled test data called fixtures, available browser actions, and assertions—exact conditions that must be true for a pass. Jev may explore, but those assertions make the final decision.
Playwright opens separate browser sessions and saves traces, which are records of actions and pages for replay. A baseline uses saved behavior without model calls; live exploration has request limits. These records help explain one run but do not prove every path works.
The project is unfinished, requires specific development tools, and currently provides no license granting use. Model exploration can miss paths, so stable checks still need fixed fixtures and exact assertions. It should not be presented as ready for adoption.
Key takeaways
- Use model judgment for exploration, not the final pass condition.
- Set shared limits on requests and processed text before live testing.
- Save enough evidence to replay and investigate each reported failure.