cyberneticlibrary

Grade adversarial test corpus with consensus

ix-adversarial-llm-panelworkflowsetup L30
GuitarAlchemist/ix
What it does

Grade the LLM-deferred tier of the ga-chatbot adversarial corpus with a multi-lens judge panel + hexavalent consensus, scored against expected_verdict

Best for

Complex workflows requiring parallel agents, synthesized judgment, or multi-phase triage.

Inputs
  • · structured data
Outputs
  • · analysis results
Requires
  • · Claude Code agent runtime (parallel/fan-out)
Preconditions
  • · Claude Code workflow harness
Failure modes
  • · Subagent produces low-quality output → iteration needed
Trust signals
  • · Adversarial cross-check phase