Are you really suggesting that the only time an "actually credible decision desk" has made a bad call was over 25 years ago?

I'm saying the fact you're resorting to that example at all is pathetic. You're digging back over a quarter-century – when models were less refined – to a call on a razor-thin race that ended up correct sans the hypothetical recount and that led to a resignation.

Of the decision desks I listed, Decision Desk HQ is almost always the quickest to call races, and even they exceedingly rarely get it wrong. The only recent one that I know of is the NJ-11 primary. They also last year miscalled the Seattle mayor's race, but that's because they received wildly inaccurate data from Seattle themselves.

Genuinely the only noteworthy one I can think of from the AP within the last 15 years is in 2018 in a House election. And errors like that (as noted in the article) get decision desks to painstakingly and methodically improve their models. The actual estimate of retracted calls:

Retractions are rare. Scott estimated the AP had to pull back about two dozen of its 7,000 calls this election season [2018], though he said most were due to county vote tabulation errors.

Unlike AlterNet, decision desks have a reputation to uphold, the expertise to uphold it, and an accountability that forces them to adapt and improve their model. Some anonymous author on AlterNet has none of those.

I'm sure if I did some research I could find plenty of retractions between then and now.

"I'm sure if I actually tried to substantiate the stupid argument I'm making with actual, salient points, I could." Well I just "did some research", and it shows pretty clearly that out of tens of thousands of races, decision desks are trustworthy.