Tangram: The Assembly Layer That Builds Search Results
Tangram evaluates nothing itself — it takes ratings from NavBoost and Glue and decides what appears on the results page, where, and in what order
What is Tangram?
Tangram is the assembly layer. It does not evaluate content quality, user satisfaction, or relevance. It takes the scores from NavBoost and Glue and builds the final search results page: which elements appear, where they appear, and in what order.
The name 'Tangram' comes from the Chinese puzzle game where seven geometric pieces are rearranged to form different shapes. The analogy is precise: the same set of search results (the pieces) can be assembled in different ways depending on the query, the user, and the context.
Tangram is fed by NavBoost (System 1) and Glue (System 2). NavBoost provides the quality and satisfaction scores for blue links. Glue provides the interaction data for special features. Tangram combines these into a single SERP layout decision.
How Tangram builds the SERP
Tangram decides which elements appear on the results page. A query for 'Hamburg weather' might get a weather carousel, a map, organic results, and a 'People also ask' box. A query for 'how to fix a leaky faucet' might get a featured snippet, video results, and organic results. Tangram makes these layout decisions based on the scores from NavBoost and Glue.
The assembly is not just about order — it is about composition. If NavBoost shows that users consistently click weather carousels for weather queries, Tangram will show weather carousels. If Glue shows that users expand 'People also ask' boxes for how-to queries, Tangram will include PAA boxes. The SERP layout is a function of user behavior data.
This creates a feedback loop. Better content in a carousel → more interaction (Glue) → Tangram shows more carousels → more opportunity for interaction. Poor content in a carousel → less interaction → Tangram shows fewer carousels. The SERP adapts to user behavior in real time.
Tangram in the ranking architecture
Tangram has no downstream dependencies (feedsInto is empty). It is a terminal system — the final step before the SERP is rendered. Everything else feeds into Tangram; nothing feeds out of it.
The E-E-A-T dimension mapping shows Tangram as indirect for both Authority (A: indirect) and Trust (T: indirect). This makes sense: Tangram doesn't measure authority or trust — it assembles results based on signals that do. But the assembly decision itself has indirect authority and trust effects: which results appear first signals which results Google trusts most.
Architecturally, Tangram is the bridge between scoring and presentation. The ranking systems (NavBoost, chard, Q*, etc.) produce scores. Tangram converts those scores into a visible SERP. Without Tangram, the scores would be numbers in a database. With Tangram, they become the search results users see.