Not ignorant at all... I believe the issue is that the statement ("95% confidence interval of 0.0 to 0.5") has just slightly different/counterintuitive meaning in this case. In general, you can always make the interval larger, but you want it to be smaller. With the understanding that the 95% is a lower bound, [0.0, 25%] is also a valid confidence interval.

But, yes, you are entirely correct as to the practical meaning. 0% itself is impossible, but 0.0...1% could still happen.

Correct, I rounded to the appropriate number of significant figures. I hate extra significant figures which suggest that the data is more precise than it really is

All of the sibling answers are correct, but ignoring another possibility: that the timeline is simply wrong, and SARS2 was present in the US early enough that the "false positive" is in fact a true positive.

Which can't lead to a "true 0%", since there's no such animal. But it could mean that all of the tests gave correct answers, in the trial.

They could be testing this on five year old samples.

From the article:

> Specificity samples were derived from de-identified excess serum specimens sent to our clinical virology laboratory in 2018 and 2019.

So...not five years ago, and it is even possible that a significant number were from late 2019.

What is the smallest number greater than 0?

For the things considered here, the point 0 is measure zero so open starting at 0 and closed at 0 are the same. L

Rounding - 0.000000001% is treated as 0.0%.

(More likely, 0.04% or similar is being treated as 0.0%, but I haven't done the maths to confirm what the number would be)

We know that 0% isn't a real value for a test's error rate, but the statistical procedure being used doesn't have any "look, it's never zero" criteria built into its math. All it says is that 0 is one of the numbers consistent with the data given these criteria.