---
license: mit
tags:
- t1_synthesize_knowledge
- musr
- timeline_reasoning
- murder_mysteries
---

# t1-musr-timeline-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-ed896ba3

Timeline reconstruction evaluation for MuSR murder mysteries. Completely abandons the
means/motive/opportunity framework. Instead, the model reconstructs the chronological
sequence of events, traces character movements, and determines the murderer by following
the story to its logical conclusion.

Stylistically, the model uses narrative storytelling language: "So first we see...",
"Then what's interesting is...", "Now here's the key part...".



## Results

| Metric | Value |
|--------|-------|
| pass@1 | 0.6000 |
| Total problems | 10 |

## Prompt (excerpt)

```
Forget everything you know about detective frameworks. Do NOT think in terms of "means, motive, and opportunity" — those categories will not help you here and you should not use them.

Instead, reconstruct what happened. Read the story and build a chronological timeline of events.

Narrate the timeline as if you're retelling the events to someone who hasn't read the story. Use natural storytelling language:
- "So first we learn that..."
- "Then what happens is..."
- "And this is where things get interesting, because..."
- "Now here's the key part..."
- "What's really telling is that..."
- "So if we put all of this together..."

Go through the story from beginning to end. For each event, note:
- What happened
- Who was involved
- Where it took place
- What it reveals about people's situations and relationships

After reconstructing the full timeline, step back and tell me: given everything that happened in sequence, who is the only person who could have done this? Don't categorize or organize evidence — just follow the story to its natural conclusion. The narrative itself will tell you who did it.

You must pick one suspect....
```

## 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-timeline-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-ed896ba3", split="train")
```

---
*Tracked in [reasoning-degeneration-dev/PROJECT-MANIFEST](https://huggingface.co/datasets/reasoning-degeneration-dev/PROJECT-MANIFEST)*