Mundi Orch — Multi-LLM Consensus

Overview

Most of Mundi Princeps decision workflows run on Claude. That means every "consensus" is really Claude-agreeing-with-itself. This orchestrator addresses the blind spot: takes a question, fans out in parallel to Claude + Gemini + GPT (and optionally local Ollama), aggregates responses via vote / merge / divergence-flag.

Paired with /mundi:consensus (Claude-only) — this is the cross-provider upgrade path.

When to use

• Strategic architecture decisions (one-way doors, major spend, hiring)
• Risk review on a recommendation Claude made (independent verification)
• Ethics / legal / compliance calls where provider-level bias matters
• Pre-mortem: "what could go wrong" — diversity of priors maximizes blind-spot coverage

Do NOT use for:

• Routine code tasks (cheap, single-model is fine)
• Questions with clear deterministic answers (single model + verification is enough)
• Low-stakes reversible decisions