What (One-liner)

Canvas single-agent LLM debate: plays a specific institution identity (CIO/quant/researcher etc.), Phase K data driving system prompt, generates {stance, reasoning, confidence, assetViews}.

How to use

Institution template (fedFOMC.member_1) → institution_agent → institution_composite
                                          → stance=bullish, confidence=0.75

Core formulas

See frontmatter. Phase K data → four-block system prompt → LLM → extract structured argument.

Assumptions & applicability

Assumptions: temp=0, maxTokens≥1024, Phase K data frozen.

Applies to: institution persona views, institution_composite children, Phase K validation demos.

Does not apply to: multi-agent consensus, byte-deterministic demos, live decisions.

Known limitations

See frontmatter. Six items: non-determinism / subjective biases / few voiceStyles / template fallback low-quality / no cross-validation / small maxTokens danger.

References

• Li et al. (2023) — LLM multi-agent debate
• Pangura Phase K Data Spec (2026-04)

Golden Test

Tier 5 attestation — pending Phase B framework.

Changelog

• 1.0.0 (2026-04-21) — First Active (Tier 5 track)