# Phase 6 confirmatory design (live OpenAI; Anthropic cells blocked by the invalid key — see docs/model_cards.md).
# Matched seeds 201–207 (7 seeds, 80 % power for paired d ≈ 1.2; all primary outcomes Holm-adjusted). Frozen 2026-10-27
# after the pilot (same code revision as the pilot except the pre-registered bug fixes recorded in docs/research_log.md).
name: confirmatory_openai
base: baseline.yaml
seeds: {start: 201, n: 7}
overrides: {population.llm.provider_mix: {openai: 1.0}, population.llm.models: {openai: gpt-5.4-nano, openai_mini: gpt-5.4-mini, mock: mock-v1, anthropic: claude-haiku-5-5},
            population.llm.usd_cap: 2.0, population.llm.concurrency: 8}
cells:
  # --- A. dose–response in the share of LLM-controlled trader slots (single generalist prompt, one model) ---
  - {name: dose_000, overrides: {population.llm_share: 0.00}, treatment: {arm: dose, share: 0.00, model: gpt-5.4-nano, prompts: single}}
  - {name: dose_005, overrides: {population.llm_share: 0.05}, treatment: {arm: dose, share: 0.05, model: gpt-5.4-nano, prompts: single}}
  - {name: dose_010, overrides: {population.llm_share: 0.10}, treatment: {arm: dose, share: 0.10, model: gpt-5.4-nano, prompts: single}}
  - {name: dose_025, overrides: {population.llm_share: 0.25}, treatment: {arm: dose, share: 0.25, model: gpt-5.4-nano, prompts: single}}
  - {name: dose_050, overrides: {population.llm_share: 0.50}, treatment: {arm: dose, share: 0.50, model: gpt-5.4-nano, prompts: single}}
  - {name: dose_100, overrides: {population.llm_share: 1.00}, treatment: {arm: dose, share: 1.00, model: gpt-5.4-nano, prompts: single}}
  # --- B. diversity at 25 %: same model + diverse persona prompts; two-model mix (nano + mini) with diverse prompts ---
  - {name: div_prompts_025, overrides: {population.llm_share: 0.25, population.llm.prompt_set: diverse}, treatment: {arm: diversity, share: 0.25, model: gpt-5.4-nano, prompts: diverse}}
  - {name: div_models_025, overrides: {population.llm_share: 0.25, population.llm.prompt_set: diverse, population.llm.provider_mix: {openai: 0.5, openai_mini: 0.5}}, treatment: {arm: diversity, share: 0.25, model: mix-nano-mini, prompts: diverse}}
  # --- C. cognition channels at 25 %: bounded memory, communication channel, adaptive P&L feedback ---
  - {name: memory_025, overrides: {population.llm_share: 0.25, population.llm.memory: 5}, treatment: {arm: cognition, share: 0.25, memory: 5}}
  - {name: channel_025, overrides: {population.llm_share: 0.25, population.llm.channel: true}, treatment: {arm: cognition, share: 0.25, channel: true}}
  - {name: adaptive_025, overrides: {population.llm_share: 0.25, population.llm.adaptive_prompt: true}, treatment: {arm: cognition, share: 0.25, adaptive: true}}
  # --- D. interaction with leverage and safeguards at 25 % ---
  - {name: leverage_025, overrides: {population.llm_share: 0.25, population.mix: {fundamentalist: 0.25, trend: 0.06, meanrev: 0.10, noise: 0.35, market_maker: 0.05, statarb: 0.0, leveraged: 0.19}}, treatment: {arm: leverage, share: 0.25, leveraged: 0.19}}
  - {name: breaker_025, overrides: {population.llm_share: 0.25, exchange.halt_threshold: 0.05}, treatment: {arm: safeguard, share: 0.25, breaker: 0.05}}
  - {name: poslimit_025, overrides: {population.llm_share: 0.25, exchange.position_limit: 300}, treatment: {arm: safeguard, share: 0.25, position_limit: 300}}
  # --- E. fairness: equal USD ≈ 25 % nano vs 5 % mini (mini costs ~5x per decision); matched latency ---
  - {name: fair_mini_005, overrides: {population.llm_share: 0.05, population.llm.provider_mix: {openai_mini: 1.0}}, treatment: {arm: fairness, share: 0.05, model: gpt-5.4-mini, prompts: single}}
  - {name: fair_latency_025, overrides: {population.llm_share: 0.25, population.latency.llm_latency: 0.05}, treatment: {arm: fairness, share: 0.25, llm_latency: 0.05}}
  # --- F. matched classical comparator for the leverage interaction (no API): same seeds, 0 % LLM, leveraged 19 % ---
  - {name: leverage_000, overrides: {population.llm_share: 0.00, population.mix: {fundamentalist: 0.25, trend: 0.06, meanrev: 0.10, noise: 0.35, market_maker: 0.05, statarb: 0.0, leveraged: 0.19}}, treatment: {arm: leverage, share: 0.00, leveraged: 0.19}}