In Giskard Hub, build datasets of scenarios with explicit checks, choosing LLM-judge, embedding, or rule-based criteria per scenario. Reference built-in checks by identifier and params, and wait for evaluation completion before reading results.

Context: Official Giskard docs (Hub SDK quickstart): documents the evaluation building blocks. A dataset is a collection of scenarios with expected outcomes and checks. Each scenario's checks field controls the criteria applied to the agent response, and checks can be LLM-judge, embedding similarity, or rule-based, referenced by identifiers like hub_correctness with params. Evaluations send every scenario to your agent and score the responses; use the wait helper to block until completion.

## Matched source
Source: Published skill
Original query: "Giskard Hub: scenarios plus checks, LLM-judge or embedding or rule-based"
Key terms: based, checks, embedding, giskard, judge, plus, rule, scenarios