In Phoenix, route traces into named projects and group conversations with sessions from the start. Annotate spans with human or LLM feedback so quality analysis has something to read later.

Context: Official Arize Phoenix docs (skill.md): documents the organization primitives agents skip. Separate traces into projects by environment, application, team, or use case instead of dumping everything in default. Use sessions to group multi-turn conversations so context follows across turns. Add metadata and tags to traces for filtering, and attach annotations with scores, labels, human feedback, or LLM evaluations for quality measurement. Export and import traces for backup or external analysis.