Short-term bridge for environments where Codex/Arize hooks already emit OpenInference spans to Phoenix. When an AgentV eval run happens inside that externally instrumented session, persist non-secret correlation metadata with the AgentV run and let the Dashboard resolve the matching Phoenix project/session.
The Dashboard should read Phoenix directly through Phoenix public GraphQL/API surface, matching how the Phoenix web UI renders sessions, traces, and spans. Do not require the px CLI at runtime, and do not query Phoenix database tables directly.
This is read-through observability, not AgentV-to-Phoenix export. AgentV artifacts, transcripts, results, and indexes remain the source of truth for the AgentV run; Phoenix remains an optional external trace database for correlation and drilldown.
Short-term bridge for environments where Codex/Arize hooks already emit OpenInference spans to Phoenix. When an AgentV eval run happens inside that externally instrumented session, persist non-secret correlation metadata with the AgentV run and let the Dashboard resolve the matching Phoenix project/session.
The Dashboard should read Phoenix directly through Phoenix public GraphQL/API surface, matching how the Phoenix web UI renders sessions, traces, and spans. Do not require the px CLI at runtime, and do not query Phoenix database tables directly.
This is read-through observability, not AgentV-to-Phoenix export. AgentV artifacts, transcripts, results, and indexes remain the source of truth for the AgentV run; Phoenix remains an optional external trace database for correlation and drilldown.