I only noticed because I ran screens today that were identical to those I ran in November, and got different results. Something must have changed to produce the difference.

I believe - but could well be wrong - that for non-standard tickers the GTR1 downloads the data (from Yahoo!?) at the time the backtest is requested and uses the downloaded data to perform the calculation. In this case non-standard means non-SI-Pro and other datasets that the GTR1 normally uses. If this is true, it's quite possible that the underlying data changed - for Yahoo! for example this happens quite frequently (and especially often for 'adjusted close' as the Dividends are recorded. I've also personally found Yahoo's calculation to be suspect on this). I believe that it was dissatisfaction with those types of datasets & the frequent errors that led to the creation of the GTR1 in the first place.

(Unless a lot of people are running the exact same request. In those instances the GTR1 can detect this and use the previously cached calculation - as it'll be the same.)

And so personally I always look at those 'downloaded' backtests with a bit of skepticism. They'll be approximately correct, but not necessarily exactly IMO.