Looking at this further, I think both I and the author of the posted tweet both initially misinterpreted what these graphs represent. The horizontal axis is the date at which the forecast (or nowcast) was made. The Atlanta Fed made estimates for Q2 GDP every day in July, which is what was reflected in the archived version of the page. They didn't make a forecast of Q3 GDP until July 30, which is why there aren't earlier data points in the current version of the graph. That's also why the latest estimate in the archived version is -1.2% while the estimate in the current version is +2% - it's not a change in estimate, they're just measuring different things. Makes perfect sense when you know what you're looking at, but easy to misinterpret without context.
It appears that nothing at all is going on. The old (“removed”) data seems to be forecasts, by date of forecast, of Q2 GDP. The new graph is forecasts of Q3 GDP. This is very cleverly hidden in the title of the graph, where no reader could possibly be expected to look. I imagine the forecasters didn’t start forecasting Q3 GDP in time for, say, an April 7 data point to ever have existed in the first place. So I would rank the surface appearance of sinisterness at approximately zero.
But maybe it is indeed even less sinister. The Atlanta Fed GDPNow site seems to have fairly extensive historical data linked near the bottom of the page.
So really the only thing to see here is that the Atlanta Fed has been reusing the same file name for updated versions of their PDF deck instead of making a new filename every week.
Live site: https://www.atlantafed.org/cqer/research/gdpnow
Latest archived version at the time of this writing: https://web.archive.org/web/20220801040302/https://www.atlan...
Latest archived version with the data in question: https://web.archive.org/web/20220729004224/https://www.atlan...