How these are built
Three rules apply to all of them.
The first is that where the evidence gives a range, you get a range. A tool that returns a single confident figure is hiding the uncertainty rather than resolving it, and the precision is usually invented.
The second is that anything contested is labelled as contested. Where the research on a claim is mixed, the mixed version is what gets published, including the studies that argue against it. The reflexes and reaction-time research behind the reflex test page is a good example — a strong correlation, a contested causal claim, and a body of work with a known publication-bias problem, all stated as such.
The third is that figures with no source do not get repeated. Several of these pages exist because a widely-copied claim turned out to have nothing behind it, and the useful thing to publish was the correction.
Each tool states what it uses and where that came from, so you can check the reasoning rather than take the output on faith.