--- license: mit tags: - t1_synthesize_knowledge - musr - structured_reasoning - murder_mysteries --- # t1-musr-structured-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-4d7e26c9 Structured slot-filling evaluation for MuSR murder mysteries. Instead of the original paper's freeform cot+ hint, this uses a structured algorithm that forces the model to extract facts, slot-fill means/motive/opportunity, and compare suspects. ## Results | Metric | Value | |--------|-------| | pass@1 | 0.8000 | | Total problems | 10 | ## Structured Prompt (excerpt) ``` You are solving a murder mystery. You must determine which suspect committed the murder. The murderer needs ALL THREE of the following established: - MEANS: Access to the murder weapon - MOTIVE: A reason to kill the victim - OPPORTUNITY: Access to the crime scene at the time of the murder Innocent suspects may have at most two of these three proven, but never all three. FOLLOW THIS EXACT STRUCTURED REASONING PROCEDURE: For EACH suspect listed in the choices below, perform this analysis: === SUSPECT: [Name] === STEP 1 — EXTRACT KEY FACTS Find exactly two key facts from the story about this suspect's relationship to the crime: Fact 1: [direct quote or close paraphrase from the story] Fact 2: [direct quote or close paraphrase from the story] STEP 2 — SLOT-FILL ANALYSIS Using ONLY the facts extracted above, determine each category: MEANS (weapon access): [ESTABLISHED / NOT ESTABLISHED] — Evidence: [cite which fact] MOTIVE (reason to kill): [ESTABLISHED / NOT ESTABLISHED] — Evidence: [cite which fact] OPPORTUNITY (scene access): [ESTABLISHED / NOT ESTABLISHED] — Evidence: [cite which fact] Verdict: [X/3 established] === END SUSPECT === After analyzing ALL suspects, build a comparison: COMPARISON: | Suspect | Means | Motive | Opportunity | Total | |---------|-------|--------|-------------|-------| (fill in for each suspect) CONCLUSION: The murderer is the suspect with ALL THREE established. If both suspects appear to have all three, pick the one whe... ``` ## Details | Parameter | Value | |-----------|-------| | Model | `together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct` | | Temperature | 0.7 | | Top-p | 0.95 | | Max tokens | 32768 | | Samples | 1 | | Thinking | False | ## Usage ```python from datasets import load_dataset dataset = load_dataset("reasoning-degeneration-dev/t1-musr-structured-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-4d7e26c9", split="train") ``` --- *Tracked in [reasoning-degeneration-dev/PROJECT-MANIFEST](https://huggingface.co/datasets/reasoning-degeneration-dev/PROJECT-MANIFEST)*