Member Junction
    Preparing search index...
    • The routing decision's input for one corpus point:

      • Message is the point's new message, and ContinuityAgentId its previous agent;
      • Participants are the history's agents, newest first, from the shared builder;
      • RecentTurns (and RecentTurnParts) come from the history through the shared builders;
      • ArtifactVersions (and ContinuityArtifacts) are the point's artifacts, as the previous agent's;
      • AllowedAgentIDs is null: the eval runs as a chat with no allowed list.

      Agents resolve through the catalog by ID, falling back to the point's own names. The history is the conversation before the new message; a history row that is the new message itself (the point's ID, or a trailing user row with the same text) is dropped, as the chat drops it.

      Parameters

      • point: {
            artifacts: {
                artifactId: string;
                artifactName: string;
                artifactType: string;
                versions: {
                    versionId: string;
                    versionName: string | null;
                    versionNumber: number;
                }[];
            }[];
            conversation_id: string;
            created_at: string;
            history: {
                agent_id: string
                | null;
                agent_name: string | null;
                created_at: string;
                id: string;
                message: string;
                role: "User" | "AI";
                status: string;
            }[];
            id: string;
            latest_message: string;
            previous_agent: { description: string; id: string; name: string };
            source: string;
        }

        The corpus point.

      • catalog: DecisionEvalAgentCatalog

        The agents to resolve against.

      Returns RoutingDecisionInput