WRT breaking down by year, you could save some download time that way. Lets say you express the predicted error in the data not as a percentage of temperature

WRT breaking down by year, you could save some download time that way. Lets say you express the predicted error in the data not as a percentage of temperature but as a time along the trendline. That would imply beyond a certain estimated error horizon, it doesn't really matter which year you download as long as you get the decade vaguely correct, once the temperature error figure exceeds a couple years of travel along the trend lines. By example lets say today, here, 2080 CE, is 35C +/- 2C (to make the math easy) but the trend line is 0.1C/yr (unrealistic, but make the math easy) then +/- 2C is the same thing as saying +/- 20 years along the trendline, so I don't really need the 2080 data set I can do "roughly as well" with any data set downloaded from 2070 to 2090. Yeah yeah I know calculus and the derivative is likely not constant or even linear and this is a really simplified way to look at statistics, but the general plan holds as a way to cut back on downloaded data required. Based on REAL statistical analysis you could come up with a formula that says you'll only increase the error bars 10% wider at year X if you skip every Y years of data. Where I'm guessing Y might exceed a decade at the extreme future years of the run. That could save an enormous amount of bandwidth without significantly impacting a visualization or analysis. "Someone else" has to download it all, run the stat analysis, then tell us all, obviously it doesn't scale for all of us to download it all, run our own stat analysis, then go back in time and only download every Y-th year at or beyond year X. So thanks in advance, "Someone else"!