A review bot's security findings are almost always accepted, but its "performance" findings are dismissed about 85% of the time. Developers have started resolving every bot comment without reading it, including valid security findings. What is the best first step?

Answer: D — Temporarily disable the performance category while its criteria are reworked

A high false positive category undermines trust in the accurate ones. Switching it off restores confidence in the security findings while its prompt is improved. A blanket "be conservative" instruction (a) doesn't target the noisy category. Confidence sorting (b) still posts the noise. Agreement across runs (c) suppresses intermittently caught real issues and leaves the category's criteria as vague as before.