SPB Git

spb/wp45_uqo

Public
776 B · 10 lines yaml
1protocol: heldout_v12split_file: data/splits/heldout_v1.json3conditions: [A_baseline, B_strong_prompt, C_conventional_rag, D_profiles_retrieval, E_full_engine]4ablations: [E_no_retrieval, E_no_profile, E_no_referee, E_no_factcheck]5tasks: [draft_introduction, rewrite_literature, explain_specification, interpret_results, improve_table, flag_causal_claims, referee_report, improve_abstract]6repetitions: 27judge: {model: claude-sonnet-5-5, blind: true, randomized_order: true, rubric: prompts/evaluation/judge_rubric.md}8primary_outcomes: [clarity, argument_logic, contribution_positioning, methodological_rigor, fidelity, usefulness]9secondary_outcomes: [unsupported_claim_count, numeric_preservation_rate, citation_verification_rate, ngram_overlap_max, latency_s, cost_usd]10
Files · main
Menu

Theme