Some Apple Watches are FDA-approved for various functions, so he is incorrect regarding data reliability. I believe it is accurate enough, as the FDA certified it for AFib, backed by several years of studies conducted by Stanford.
Likewise, he quotes oxygen measurements of 88% to 97% -- the FDA standard is that 95% of SpO2 values fall within 4 to 6% of the true reading. So this type of variation is expected and part of why you should look at trends rather than individual outliers.