How the ranking order is determined

German law (§ 5b (2) UWG) requires every result list to disclose how its order comes about. This page shows the actually active configuration live from the database — never hardcoded.

Two stages — no promise that a higher score means a better position

1. 1. Scoring: each result gets a score made of two parts — retrieval relevance from the candidate search and the share of met requirements, weighted according to the active configuration shown below.
2. 2. Reranking: step 1 determines the starting order. A downstream step (text similarity between the query and the model description, BM25) then reorders this list once it finds its own signal — without such a signal (e.g. when none of the query's words appear in the model text), the order from step 1 is kept. A higher score therefore does not necessarily mean a better position.

For a structured recommendation ("Find the best AI"), step 2 is skipped entirely — there is no free text against which retrieval relevance could be measured. Its ranking comes solely from the share of met requirements (step 1), against the same configuration shown here.

No active ranking configuration is on record right now.