I think we'll see AZ projected by Sunday Night. Here's why...
I've been puzzled by the vote counting in AZ because it seems never to change as I watch the estimated number of ballots left on fivethirtyeight's live blog thread. So I decided to make a simple linear model and try to estimate how fast the votes are being counted. This assumes that the rate has been constant and is not slowing down (which may not be true) and it assumes that more ballots won't appear from any new sources (which may also not be true). I am also assuming the networks will call the race when the number of uncounted ballots is close to zero.
Method:
I plotted the number of ballots estimated to be remaining against the number of hours which had passed since election day ended (12am wednesday). I then fit the data with a simple linear model. I obtained the equation y= -6896x + 653959 . The coefficient of determination of 0.81. This means that according to my estimates 6896 ballots are counted in AZ in an average hour, and there were 653959 uncounted ballots at 12 am 11/4/20. It also means that the correlation is not stellar.
Results:
According to my model the number of uncounted ballots will drop to 0 around midnight tomorrow (Sunday at 12 am - just before the 96 hour mark).
There's my back of the envelope calculation. Feel free to let me know about any source of error I might not have thought of or read about.