Found something similar reviewing 1,000s of bug write-ups created by Sherlock AI. AI is surprisingly good at finding rounding errors. It's one of the exploit classes where it excels. But it only finds the "Low" rounding error, rarely can it turn it into a "Critical" exploit.