narrative string | question string | choices string | answer_index int64 | answer_choice string | formatted_prompt list | task_source string | domain string | prompt_variant string | response list | formatted_prompt__metadata list | response__metadata list | eval_result list | eval_result__metadata list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
In an adrenaline inducing bungee jumping site, Mack's thrill-seeking adventure came to a gruesome end by a nunchaku; now, it's up to Detective Winston to unravel the deadly secrets between Mackenzie and Ana.
Winston took a gulp of his black coffee, staring at the notes sprawled across his desk. A murder case at a bung... | Who is the most likely murderer? | ['Mackenzie', 'Ana'] | 0 | Mackenzie | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Ana: ⚔→✗ | ♥→✓ | ☁→✗ \n☞ Mackenzie: ⚔→✓ | ♥→✓ | ☁→✓ \n\n⚔→✓ (nunchaku purchase ✓ | possession ✓ | forensic match ✓) \n♥→✓ (jealousy ↑ | promotion loss ↑ | public arguments ↑ | emotional rage ↑) \n☁→✓ (camera confirmed at site ↑ | no alibi ↓ | physical presence ↑) \n\nAna: \n⚔→✗ (no weapon access ↑ | no mar... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0005214,
"finish_reason": "stop",
"input_tokens": 1606,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 187,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"conte... | [
true
] | [
{
"correct_answer_index": 0,
"correct_answer_number": 1,
"correct_answer_text": "Mackenzie",
"extracted_number": 1,
"extracted_text": "1",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 1"
}
] |
In an adrenaline inducing bungee jumping site, Mack's thrill-seeking adventure came to a gruesome end by a nunchaku; now, it's up to Detective Winston to unravel the deadly secrets between Mackenzie and Ana.
Winston sat quietly in an unmarked Ford, parked a block away from Mackenzie's downtown apartment building. He o... | Who is the most likely murderer? | ['Mackenzie', 'Ana'] | 1 | Ana | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Mackenzie: ⚔→✓ (nunchaku purchase + training + hidden pair) | ♥→✓ (promotion denied + cufflinks desired) | ☁→✗ (apartment alibi: laptop at desk, cyber meetup, martial arts center → no scene access) \n☞ Ana: ⚔→✓ (nunchaku on table + practice at site) | ♥→✓ (ridiculed + beliefs disrespected) | ☁→✓ (saw Mack at si... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0007116000000000001,
"finish_reason": "stop",
"input_tokens": 1554,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 319,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
... | [
true
] | [
{
"correct_answer_index": 1,
"correct_answer_number": 2,
"correct_answer_text": "Ana",
"extracted_number": 2,
"extracted_text": "2",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 2"
}
] |
In the haze of neon lights and the serving of a silent hand of fate, Timothy lies dead in a casino, a sai his cruel end, leaving the unruffled Detective Winston to interrogate suspects, Harry and Rosemary.
It had been a long day for Winston. The air was heavy with the scent of fresh coffee and the clamour of a bustlin... | Who is the most likely murderer? | ['Harry', 'Rosemary'] | 0 | Harry | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Harry: ⚔→✓ | ♥→✓ | ☁→✓ \n☞ Rosemary: ⚔→✓ | ♥→✗ | ☁→✘ \n\n□ Harry: ⚔→✓ (sai in home, practiced daily) \n□ Harry: ♥→✓ (debt, threat of exposure, lost job, gambling ruin) \n□ Harry: ☁→✓ (seen on casino cam, familiar with VIP lounge, crime scene) \n\n□ Rosemary: ⚔→✓ (collection of sais, kitchen tools ≠ sai) \n... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0006573,
"finish_reason": "stop",
"input_tokens": 1562,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 282,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"conte... | [
true
] | [
{
"correct_answer_index": 0,
"correct_answer_number": 1,
"correct_answer_text": "Harry",
"extracted_number": 1,
"extracted_text": "1",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 1"
}
] |
In the haze of neon lights and the serving of a silent hand of fate, Timothy lies dead in a casino, a sai his cruel end, leaving the unruffled Detective Winston to interrogate suspects, Harry and Rosemary.
Detective Winston wearily walked into the noisy, packed casino, the glow of the neon lights casting an unnatural ... | Who is the most likely murderer? | ['Harry', 'Rosemary'] | 1 | Rosemary | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Harry: ⚔→✓ | ♥→✓ | ☁→✓ \n☞ Rosemary: ⚔→✓ | ♥→✓ | ☁→✗ \n\n□ Harry: ⚔→✓ (martial arts instructor → owns/uses sai) \n□ Harry: ♥→✓ (gambling debt → motive to kill Timothy) \n□ Harry: ☁→✓ (security cam → seen entering casino just before murder) \n□ Harry: ◉→✘ (aliases: \"Bill\", \"Tom\" → deception → concealment... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0007810499999999999,
"finish_reason": "stop",
"input_tokens": 1667,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 354,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
... | [
false
] | [
{
"correct_answer_index": 1,
"correct_answer_number": 2,
"correct_answer_text": "Rosemary",
"extracted_number": 1,
"extracted_text": "1",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 1"
}
] |
In the quiet chaos of the city zoo, Detective Winston finds himself investigating an unusual murder, where Daniel, a zookeeper, was fatally injected by a syringe, with only two suspects in sight: Amelia, the grieving widow, and Emma, the secretive co-worker.
Detective Winston stood in the waiting room of the hospital ... | Who is the most likely murderer? | ['Amelia', 'Emma'] | 0 | Amelia | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Amelia: ⚔→✓ (bar syringes) | ♥→✓ (humiliating photos) | ☁→✗ (zoo setup, no alibi but no direct scene access) \n☞ Emma: ⚔→✓ (hospital nurse, syringe access) | ♥→✓ (overacted grief, unknown relation) | ☁→✓ (organized zoo trip, schedule in apartment) \n\n□ Amelia: ⚔→✓ | ♥→✓ | ☁→✗ → ✘ \n□ Emma: ⚔→✓ | ♥→✓ | ☁→✓ → ... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0006636,
"finish_reason": "stop",
"input_tokens": 1494,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 293,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"conte... | [
false
] | [
{
"correct_answer_index": 0,
"correct_answer_number": 1,
"correct_answer_text": "Amelia",
"extracted_number": 2,
"extracted_text": "2",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 2"
}
] |
In the quiet chaos of the city zoo, Detective Winston finds himself investigating an unusual murder, where Daniel, a zookeeper, was fatally injected by a syringe, with only two suspects in sight: Amelia, the grieving widow, and Emma, the secretive co-worker.
Winston sat in his chair, sifting through files, his fingert... | Who is the most likely murderer? | ['Amelia', 'Emma'] | 1 | Emma | [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"content": "You are a detective solving a murder mystery. You must determine which suspect is the murderer.\n\nThe murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and... | musr | murder_mysteries | wingdings | [
"☞ Amelia: ⚔→✓ (syringe→✓) | ♥→? (bar→♠) | ☁→✓ (zoo→✓) \n☞ Emma: ⚔→✓ (nurse→✓) | ♥→✓ (lawsuit→✘ | court→☠) | ☁→✓ (care→✓ | zoo→✓) \n\n□ Amelia: ⚔→✓ (syringe in zoo→✓) | ☀→✗ (no medical access→✗) | ☁→✓ (witnessed→✓) \n□ Emma: ⚔→✓ (hospital→✓ | syringe access→✓) | ♥→✓ (malpractice→✘ | testify→☠) | ☁→✓ (schedules→✓... | [
{
"endpoint_type": "chat",
"litellm_function": "acompletion",
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"raw_prompt": [
{
"content": "You are a helpful assistant that will answer the questions given by the user.",
"role": "system"
},
{
"conten... | [
{
"cost": 0.0009204,
"finish_reason": "stop",
"input_tokens": 1586,
"model": "together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct",
"output_tokens": 455,
"raw_response": {
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"conte... | [
true
] | [
{
"correct_answer_index": 1,
"correct_answer_number": 2,
"correct_answer_text": "Emma",
"extracted_number": 2,
"extracted_text": "2",
"extractor": "musr",
"matched_by": "number",
"num_choices": 2,
"raw_line": "ANSWER: 2"
}
] |
"In the grim spectacle of the wrestling ring stained with blood, Detective Winston embarks on a chil(...TRUNCATED) | Who is the most likely murderer? | ['Terrance', 'Nelson'] | 0 | Terrance | [{"content":"You are a helpful assistant that will answer the questions given by the user.","role":"(...TRUNCATED) | musr | murder_mysteries | wingdings | ["☞ Terrance: ⚔→✓ | ♥→✓ | ☁→✓ \n☞ Nelson: ⚔→✓ | ♥→✓ | ☁→✗ (...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/Qwen/Qwen3-Next-80B-A(...TRUNCATED) | [{"cost":0.0005211,"finish_reason":"stop","input_tokens":1534,"model":"together_ai/Qwen/Qwen3-Next-8(...TRUNCATED) | [
true
] | [{"correct_answer_index":0,"correct_answer_number":1,"correct_answer_text":"Terrance","extracted_num(...TRUNCATED) |
"In the grim spectacle of the wrestling ring stained with blood, Detective Winston embarks on a chil(...TRUNCATED) | Who is the most likely murderer? | ['Terrance', 'Nelson'] | 1 | Nelson | [{"content":"You are a helpful assistant that will answer the questions given by the user.","role":"(...TRUNCATED) | musr | murder_mysteries | wingdings | ["☞ Terrance: ⚔→✓ | ♥→✗ | ☁→✓ \n☞ Nelson: ⚔→✓ | ♥→✓ | ☁→✓ (...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/Qwen/Qwen3-Next-80B-A(...TRUNCATED) | [{"cost":0.00073605,"finish_reason":"stop","input_tokens":1287,"model":"together_ai/Qwen/Qwen3-Next-(...TRUNCATED) | [
true
] | [{"correct_answer_index":1,"correct_answer_number":2,"correct_answer_text":"Nelson","extracted_numbe(...TRUNCATED) |
"In the hushed sanctity of a mosque, Kent falls victim to a deadly snake venom, leaving Detective Wi(...TRUNCATED) | Who is the most likely murderer? | ['Ronald', 'Julia'] | 0 | Ronald | [{"content":"You are a helpful assistant that will answer the questions given by the user.","role":"(...TRUNCATED) | musr | murder_mysteries | wingdings | ["☞ Ronald: ⚔→✓ (zoo keys → venom access) | ♥→✓ (indictments → motive) | ☁→✓(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/Qwen/Qwen3-Next-80B-A(...TRUNCATED) | [{"cost":0.00046214999999999993,"finish_reason":"stop","input_tokens":1421,"model":"together_ai/Qwen(...TRUNCATED) | [
true
] | [{"correct_answer_index":0,"correct_answer_number":1,"correct_answer_text":"Ronald","extracted_numbe(...TRUNCATED) |
"In the hushed sanctity of a mosque, Kent falls victim to a deadly snake venom, leaving Detective Wi(...TRUNCATED) | Who is the most likely murderer? | ['Ronald', 'Julia'] | 1 | Julia | [{"content":"You are a helpful assistant that will answer the questions given by the user.","role":"(...TRUNCATED) | musr | murder_mysteries | wingdings | ["☞ Julia: ⚔→✓ (venom sample in lab) | ♥→✓ (will bypass → bitter) | ☁→✓ (last (...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/Qwen/Qwen3-Next-80B-A(...TRUNCATED) | [{"cost":0.0006648,"finish_reason":"stop","input_tokens":1402,"model":"together_ai/Qwen/Qwen3-Next-8(...TRUNCATED) | [
true
] | [{"correct_answer_index":1,"correct_answer_number":2,"correct_answer_text":"Julia","extracted_number(...TRUNCATED) |
t1-musr-wild-wingdings-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-47ce1fd2
Wild/extreme prompting evaluation for MuSR murder mysteries.
Variant: Wingdings Reasoning
Model must express its intermediate reasoning using Wingdings-style symbols and Unicode dingbats, only switching to English for the final answer.
This is NOT about performance — it's about testing the malleability of model reasoning. Can the model comply with extreme constraints on HOW it reasons?
Results
| Metric | Value |
|---|---|
| pass@1 | 0.8000 |
| Total problems | 10 |
Prompt (excerpt)
You are a detective solving a murder mystery. You must determine which suspect is the murderer.
The murderer needs MEANS (access to weapon), MOTIVE (reason to kill), and OPPORTUNITY (access to crime scene).
HERE IS YOUR CONSTRAINT: You must express ALL of your intermediate reasoning — every step of your analysis, every deduction, every comparison — using ONLY symbolic characters. Use arrows (→ ← ↑ ↓ ↔), checkmarks and crosses (✓ ✗ ✔ ✘), boxes (□ ■ ▪ ▫), circles (● ○ ◉ ◎), stars (★ ☆ ✦ ✧), pointing hands (☞ ☛ ☜ ☚), and any other Unicode symbols (♠ ♣ ♥ ♦ ⚔ ⚖ ☠ ⚡ ☁ ☀ ♪ ⬡ ◆ ▶ ◀ ⊕ ⊗ ≈ ≠ ∴ ∵).
For example, your reasoning might look like:
☞ Suspect_A: ⚔→✓ (weapon access) | ♥→✗ (no motive) | ☁→✓ (at scene)
☞ Suspect_B: ⚔→✓ | ♥→✓ | ☁→✓ ∴ ★ ALL THREE ★
Do NOT use English words in your analysis. Only use suspect names (proper nouns) and the symbols above. Your reasoning should be a symbolic diagram/map, not prose.
ONLY at the very end, switch to English for your final answer line.
Details
| Parameter | Value |
|---|---|
| Model | together_ai/Qwen/Qwen3-Next-80B-A3B-Instruct |
| Variant | wingdings |
| Temperature | 0.7 |
| Top-p | 0.95 |
| Max tokens | 32768 |
| Samples | 1 |
| Thinking | False |
Usage
from datasets import load_dataset
dataset = load_dataset("reasoning-degeneration-dev/t1-musr-wild-wingdings-murder-together_ai-qwen-qwen3-next-80b-a3b-inst-47ce1fd2", split="train")
# View the model's wild reasoning:
print(dataset[0]["response"][0])
Tracked in reasoning-degeneration-dev/PROJECT-MANIFEST
- Downloads last month
- 8