For a brief moment, we knew how many Steam games had sold

Unlike many other sectors of the entertainment industry, PC gaming is remarkably opaque with respect to sales data. Valve Software's Steam market is an important part of all PC game sales, but historically the data has been difficult or impossible to see and even more difficult to independently verify.

For the shortest moment of this week, we had a fabulously clear snapshot of the number of games that Steam actually sold. We still have the picture, but it is the last we will have soon. Just days after opening a window on their organization, Valve closed it.

Here's how everything went.

The story begins in 2014 when Kyle Orland of Ars Technica performed an experiment. He sampled the stream of public, vaguely anonymous data that Valve sent about his millions of users. He called it the "Steam Gauge" and wrote about his methodology in a long feature article.

Blogger and podcaster Sergey Galyonkin, who now works for Epic Games, has read the story of Orland and set out to create a website to make this data available to everyone. It was called Steam Spy, and Polygon first talked about it in 2015. For about three years, the public had access to a steady stream of player accounts and sales history for each game on Steam. Galyonkin himself became a kind of minor celebrity, even delivering a lecture at the Game Developers Conference about what he had learned by analyzing the data.

In April of this year everything changed. Valve announced that it had changed the way it disseminated user data. Many assumed that the decision was made to comply with the European Union General Protection Data Requirements (GDPR), which became enforceable in May of this year. Regardless of their reasoning, Galyonkin's service was much less precise than in the past.

In response, an enterprising programmer and game designer named Tyler Glaiel proposed a new method. With the user data flow that Valve traditionally shared, he assumed that you could use the data of a game to do pretty much the same thing. If a given percentage of owners of a given game had unlocked a particular achievement, you could work backwards and discover the number of people who owned the game.

On average, Glaiel explained his process:

This was evoked a group of developers in which I am, and it was quickly pointed out that if you get success data via the steam API, you get 16 digits of accuracy instead! I started trying to reproduce the barter.vg algorithm based on the description of it on their site, "Calculated by finding the smallest number of players producing whole numbers of players for each achievement (percentage achieved * all players) "

So I made it work, with a simple brute force. Check each possible whole number of sales up to a ceiling, and multiplied by the percentages of achievement. None of them has reached exactly an integer, so I had to define a threshold for what counts as a "whole number". It has worked for most games with less than a million sales.

With more refinement, Glaiel was able to provide data that was more accurate than ever Steam Spy. Thus, Galyonkin ran its servers and went to work using this new method. He even worked with Ars Technica to check his work. The result is a massive CSV file containing sales data for every game on Steam that delivers achievements to its users.

But this is the last time the public will be able to use this particular feat. Just a few days ago, Valve started rounding success data to the nearest whole number. Without data with 16 precision digits after the decimal point, Glaiel's method breaks down.

For a list of the 1000 best-selling games on Steam from this week, generated by Galyonkin and using Glaiel's new method, look at Ars Technica .