I'm no apologist for WU, but let's look at some numbers.
https://www.wunderground.com/weatherstation/overview.asp claims >250,000 PWSes uploading to them.
https://www.wunderground.com/webcams/ says 4,342,482 images uploaded today, 452,220 videos created today. It's a little after noon Central Time so double those numbers to get a 24 hour estimate. Almost 9 million images and 900,000 videos a day.
It's a wonder that any of it works at all.
Now then, if just 0.1% of their claimed PWS owners send an email on a given day, that's 250 emails to sort through and answer, every day.
Finally, according to
http://www.wxforum.net/index.php?action=mlist;sort=registered;start=0 wxforum has 12,226 members registered today. IF every every registered member was sending data to WU, AND we all decided to stop sending data today, they would lose 4.9% of their PWSes. I doubt they'd even notice.
We all know WU was bought by IBM. (The Weather Channel is apparently NOT owned by IBM.)
https://www.wunderground.com/blog/JeffMasters/weather-underground-bought-by-ibm.html October 28, 2015
So, IBM gets all of the live and historical WU data to play with. I doubt they care a lot about the crap data that some stations send. It's free to them, and they can filter that out easily enough, and then mash the rest, crunch it, and resell it. All they have to do is keep the masses happy enough to keep sending data. And, they cut back on what they perceive to be non-essential services, features and expenses.
(A couple of weeks ago, I had a request from a WeatherElement user to send his data to WU. It's easy enough, we do it from the server, so it's just an entry in a database table. But I asked why. It turns out the a bunch of his friends work for IBM and were hassling him about it. Peer pressure... Sheesh. So I set him up to send to PWSweather too.)
I'll throw out one more thought... As near as I can tell, the infamous "gold stars" started appearing sometime around early August 2016. This MIGHT be something some IBM programmer cooked up as a first, albeit ineffective, cut at data quality analysis, less than a year after the purchase.