Skip to content

resil-redundant-voter

Domain: resil · Model class: cheap

Use when a user wants to reduce hallucination rates or add fault-tolerance by running a node multiple times and voting on the result. Triggers: “make this more reliable”, “reduce hallucinations”, “run N times and pick the best”, “Byzantine fault tolerance”, “majority vote on agent output”, “ISS-style redundancy”, “N-modular redundancy”, outputs are “inconsistent” or “sometimes wrong” and the user wants a structural fix not a prompt tweak.

ISS quad-processor voting applied to LLM nodes. Run N identical sub-prompts in parallel (temperature-jittered), collect outputs, compute pairwise similarity, return the majority-cluster centroid.

  • “make this more reliable”
  • “reduce hallucinations”
  • “run N times and pick the best”
  • “Byzantine fault tolerance”
  • “majority vote on agent output”
  • “ISS-style redundancy”
  • “N-modular redundancy”
  • “outputs are inconsistent”
  • “sometimes wrong”
  • the user wants a one-off improvement without ongoing adaptation or structural change
  1. What node or agent output needs fault-tolerance?
  2. What n_replicas and similarity_threshold are appropriate?
  3. What tiebreak strategy should apply (escalate, abstain, longest)?
  4. What comparison mode is required (semantic, structural, exact)?
  • failure mode analysis
  • recovery strategy
  • operational checks
  • validation notes

flow-orchestrator · orch-agent-orchestrator · prompt-chaining