Brief, mostly facts
ReqTestAgent generates acceptance tests that catch real bugs
A Frontiers study tests generated acceptance tests against a reference implementation. It evaluated 100 requirement instances from 25 tasks.
What the research found
A Frontiers in Artificial Intelligence study presents ReqTestAgent, which generates acceptance tests from requirements and checks them against a reference implementation before scoring mutation strength, evaluated on 100 requirement instances from 25 Orchid and BigCodeBench tasks.
Claims and how well the evidence supports them
- Generated tests must pass a canonical reference implementation before mutation scores are interpreted 98%
- The study used 100 requirement instances from 25 tasks drawn from Orchid and BigCodeBench 99%
Evidence
- bioengineer.org https://bioengineer.org/ai-agents-learn-to-write-acceptance-tests-that-actually-catch-software-bugs
Actionability rated low.