Track the same thing on my end. It's always the cheap inlet valve or a bad float switch, never the pricey boards. People hate hearing the fix is 20 bucks in parts plus labor. Boring parts fail way more than fancy ones, every single time.

2,000 is a big number, sure, but it sounds like you're grading the tool on a curve. A 3% false positive rate means 60 requests were still garbage. If those were the complex ones you mentioned, that's a lot of manual cleanup that the counter doesn't show. Milestones like this just mean the system didn't crash, not that it's actually good. Give it another 2,000 and see if the rate holds when the contract language gets weird. This reads like you needed a reason to stop doubting it, not like the data changed your mind.

Wait, so you tracked those 5+ touchpoints yet?