02 — Compare
Most recommendation systems compare what people played, or match genre and tag overlap. Strata's similarity engine compares the actual 21-trait construction of two games and tells you exactly which traits they share.
The problem
Collaborative-filtering and tag-based recommendation systems are popular precisely because they're easy — not because they're accurate at the level of felt experience. They can surface a match because of marketing overlap, release timing, or shared platform, with zero relationship to how the games actually feel to play.
The mechanism
Every classified game has a 21-dimensional trait vector. Comparing two games means comparing those vectors directly — not metadata, not tags, not user behavior.
01
Vector built
Each game's 21 trait scores become a vector. Missing traits are treated as zero so partial data never breaks a comparison.
02
Distance calculated
Similarity is the normalized inverse of the total absolute difference across all 21 traits — closer construction means a higher score.
03
Shared anchors identified
Any trait where both games score 4 or higher is flagged as a shared anchor — the specific reason the match exists.
04
Shared lows identified
Any trait where both games score 2 or lower is flagged as a shared low — what both games deliberately chose not to build.
Proof
Death Stranding and Elden Ring are not similar on the surface — one is a delivery-and-traversal game, the other is a brutal action RPG. But both score exceptionally on Vulnerability, Isolation, and Environmental Storytelling.
That's not a coincidence the engine stumbled into. It's the similarity calculation doing exactly what it's built to do: finding the traits where two very different-looking games were actually constructed the same way.
Continue exploring