[M3] Agent Evaluation Dataset + Tracing setup

[ComBba]

Overview

Set up Gradient Evaluations with test cases and enable Tracing for pipeline debugging.

Acceptance Criteria

• Evaluation dataset (CSV) with 6+ test cases:
  • Simple CRUD app → expected GO
  • Stock trading AI → expected NO-GO
  • Portfolio website → expected GO
  • Video conferencing → expected CONDITIONAL
  • Blockchain voting → expected NO-GO
  • Recipe community → expected GO
• Evaluations configured in Gradient console
• Tracing enabled (automatic via LangGraph nodes)
• Token usage tracked per run
• Cost tracking per run

Dependencies

• [M2] LangGraph StateGraph: wire all nodes + SSE streaming entrypoint #15 (complete pipeline to evaluate)

Reference

• `docs/reference/10-technical-plan.md` — Agent Evaluation Dataset
• `docs/reference/05-digitalocean-gradient-ai.md` — Evaluations, Tracing