Wow, over twenty years on I am, belatedly, redirecting you to https://shiny.psyctc.org/apps/RCI1/. That actually replaced the defunct perl cgi-lib app some way back and I fixed the parent page to redirect things but a kind person has just alerted me that I hadn't redirected this page as I assumed people wouldn't get to it: wrong Christopher, the internet has a bigger memory than any elephant!
You might also want to look at https://shiny.psyctc.org/apps/RCI2/ That gives a "plausible interval" around the point RCI value. That is roughly equivalent to a confidence interval and is based on the standard error of the standard deviation. It can be a very useful cautionary note about how imprecisely the RCI is likely to have been estimated: not very precisely unless your n is very large.
SD1*sqrt(2)*sqrt(1-rel)where SD1 is the initial standard deviation
The criterion that is always used is the 95% criterion, i.e. a value such that change greater than this, improvement and deterioration, would only occur by unreliability of measurement alone in less than 5% of times that two measurements are made on the same person.
The following form calculates this for you: