QAVE Docs

Agent Trajectories

Neuro-symbolic tool-call judging (D-067).

Supports OpenAI / Anthropic / LangGraph tool call shapes with:

  • Retry recovery exemption (failed tool → same-tool retry success)
  • first_error_turn
  • agent_diagnostics (unrecovered error turns, orphan tools)

ScorePanel shows Agent trajectory breakdown only when agent_diagnostics != null.

Voice Assurance

Voice Mode 2 dual scores, root-cause layers, and webhook normalization.

Insights & MCP

Causal failure clustering, tag slices, and the read-only MCP server.