Steam Spy operated by using the Steam API. This allowed the website to collect information by constantly seeing which games have public Steam accounts in their possession. This way a good picture can be formed of how much a certain game is played, and with some corrections how often a game is sold.
It all worked thanks to the information provided by Steam itself. And now, as Valve does more often, that is adjusted without any announcement. The motives of Valve are therefore unclear. Perhaps the new European privacy legislation has to do with it, but as the maker of Steam Spy correctly indicates other information should be hidden.
At first, it seemed like the end from Steam Spy, but now a new way has been found. With the help of machine learning, Steam Spy can get a second life. However, that is not as obvious as it seems. Machine learning is very good at dividing things into categories, but the so-called regression problems, as is the case here, are a lot more difficult.
The first results also show large error margins. Of 70 games that have been tested, 90% fall within a margin of error of 10%. However, that is a small control group within the thousands of games on Steam, because most game developers do not reveal how often their game has actually been sold.
It is not perfect, and probably will not be in the near future. But it is more than nothing, and in an industry that is often mysterious about its sales numbers, this is more than welcome.