experiment_id: E1_pilot
splits: [dev]
families: [synthetic, exec, public_qasper, public_2wikimqa]
lengths: [2000, 8000]
limit_docs_per_family: 2
max_questions_per_doc: 5
model_pairs:
  - {sender: qwen3b, receiver: qwen3b}
  - {sender: llama3b, receiver: llama3b}
generation: {max_tokens: 48, per_family: {exec: 320}}
seeds: [0]
conditions:
  - {method: A0}
  - {method: A1}
  - {method: A2, budget_tokens: 128}
  - {method: A2, budget_tokens: 512}
  - {method: A3, budget_tokens: 128, access: blind}
  - {method: A3, budget_tokens: 512, access: blind}
  - {method: A3, budget_tokens: 512, access: aware}
  - {method: A4, budget_tokens: 128, access: blind}
  - {method: A4, budget_tokens: 512, access: blind}
  - {method: A4, budget_tokens: 512, access: aware}
  - {method: A5, budget_tokens: 128, access: blind}
  - {method: A5, budget_tokens: 512, access: blind}
  - {method: A5, budget_tokens: 512, access: aware}
  - {method: A6, budget_tokens: 512, access: blind, max_fetch_bytes: 2000, max_lookups: 6}
  - {method: A7, codec: zstd}
  - {method: A8, budget_tokens: 512, access: blind}
  - {method: B0}
  - {method: B1}
  - {method: B2, bits: 8}
  - {method: B2, bits: 4}
  - {method: B2, bits: 2, group_size: 32}
  - {method: B3, keep_frac: 0.5}
  - {method: B3, keep_frac: 0.25}
  - {method: B3, keep_frac: 0.125}
  - {method: B4, keep_frac: 0.25}
  - {method: B5, keep_frac: 0.5, access: aware}
  - {method: B5, keep_frac: 0.25, access: aware}
  - {method: B5, keep_frac: 0.125, access: aware}
  - {method: B5, keep_frac: 0.25, access: blind}
  - {method: B6, keep_frac: 0.25, access: aware}
  - {method: B9, keep_frac: 0.25, bits: 8, access: aware, policy: attention}