Steam vs Metacritic: two numbers, two different questions

Terraria: a checkable example

A positive recommendation can mean anything from a qualified yes to a favourite game. That binary choice cannot reveal the strength of every opinion. Nor does it tell us whether someone with your preferences will enjoy the game.

Why we do not calculate a critic-player gap

Subtracting a Metascore from a recommendation percentage produces arithmetic, not a shared measure of quality. The audiences, scales, editions and collection dates may differ. A big-looking difference is a question to investigate, not proof that critics or players got it wrong.

Even ranking a game separately in two lists needs a stated comparison group. A percentile describes position within that group, not the size of a disagreement.

Four checks before choosing a game

1. Check the edition and platform. A console launch review may not describe today's PC build.
2. Check the date and filters. Recent reviews and lifetime reviews cover different periods; language and purchase filters can change the included audience.
3. Read the reasons. Look for the mechanics, pacing, accessibility or co-op experience that matter to you. A sample of comments is not representative of every voter.
4. Verify practical details. Use the current store or developer documentation for price, requirements and supported features. A reviewer's anecdote is not a specification.

What an update can and cannot show

We now store dated source observations separately from AI summaries. One observation establishes a baseline, not a trend. Comparable observations on later dates can show how the recorded number changed, but cannot by themselves explain why it changed.

We keep an older analysis with its original date instead of silently presenting it as a new review. When evidence is unavailable, we say so.