id
stringlengths 1
12
| question
stringlengths 22
566
| golden_answers
dict | data_source
stringclasses 2
values | prompt
listlengths 1
1
| ability
stringclasses 6
values | reward_model
dict | extra_info
dict | metadata
dict |
|---|---|---|---|---|---|---|---|---|
exfever_7771
|
Bernard Sanders is an American politician and activist who has served as the junior United States senator from a state in the New England region of the United States.
|
{
"evidence": [
"Bernie Sanders",
"Vermont"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBernard Sanders is an American politician and activist who has served as the junior United States senator from a state in the New England region of the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bernie Sanders",
"Vermont"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 8184,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4310
|
Saddam Hussein was an Iraqi politician who served as the country's fifth president and was a leading role in the revolutionary Arab Socialist Ba'ath Party, which supported Ba'athism, a mix of Arab socialism and nationalist ideology that affirms the Arabs as a nation and advocates Arab unity.
|
{
"evidence": [
"Saddam Hussein",
"Arab nationalism"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSaddam Hussein was an Iraqi politician who served as the country's fifth president and was a leading role in the revolutionary Arab Socialist Ba'ath Party, which supported Ba'athism, a mix of Arab socialism and nationalist ideology that affirms the Arabs as a nation and advocates Arab unity.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Saddam Hussein",
"Arab nationalism"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 8185,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2747
|
21 Jump Street is a 2012 American buddy cop action comedy film adapted from the television series of the same name by an American television producer and writer.
|
{
"evidence": [
"21 Jump Street (film)",
"Patrick Hasburgh"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n21 Jump Street is a 2012 American buddy cop action comedy film adapted from the television series of the same name by an American television producer and writer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"21 Jump Street (film)",
"Patrick Hasburgh"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 8186,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8667
|
Nine Inch Nail's dark, even sadistic themes, stood in stark contrast to things like the Spice Girls and Britney Spears, who were also popular at the time.
|
{
"evidence": [
"Year Zero (album)",
"Concept album"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNine Inch Nail's dark, even sadistic themes, stood in stark contrast to things like the Spice Girls and Britney Spears, who were also popular at the time.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Year Zero (album)",
"Concept album"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 8187,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6257
|
Neymar is a Brazilian professional footballer who plays for Paris Saint-Germain as a player who plays nearest to the opposing team's goal, in a team sport played with a spherical ball between two teams of 14 players.
|
{
"evidence": [
"Neymar",
"Forward (association football)",
"Association football"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNeymar is a Brazilian professional footballer who plays for Paris Saint-Germain as a player who plays nearest to the opposing team's goal, in a team sport played with a spherical ball between two teams of 14 players.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Neymar",
"Forward (association football)",
"Association football"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 8188,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29978
|
After the concurrency Highway 2 continues northward, meeting US-270 at Wilburton; SH-2 then passes Robbers Cave State Park and the eastern edge of the Sansbois Mountains before reaching SH-31 east of Quinton.
|
{
"evidence": [
"Oklahoma State Highway 2_sentence_14",
"Oklahoma State Highway 2_sentence_15"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAfter the concurrency Highway 2 continues northward, meeting US-270 at Wilburton; SH-2 then passes Robbers Cave State Park and the eastern edge of the Sansbois Mountains before reaching SH-31 east of Quinton.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Oklahoma State Highway 2_sentence_14",
"Oklahoma State Highway 2_sentence_15"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 8189,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.774",
"value": "?search="
},
{
"operation": "search",
"time": "30.339",
"value": "SH-2"
},
{
"operation": "Now on",
"time": "31.299",
"value": "Kaman SH-2 Seasprite"
},
{
"operation": "hyperlink",
"time": "37.01",
"value": "SH2 (disambiguation)"
},
{
"operation": "Now on",
"time": "37.428",
"value": "SH2"
},
{
"operation": "hyperlink",
"time": "47.121",
"value": "List of highways numbered 2"
},
{
"operation": "Now on",
"time": "47.999",
"value": "List of highways numbered 2"
},
{
"operation": "hyperlink",
"time": "85.51",
"value": "Oklahoma State Highway 2"
},
{
"operation": "Now on",
"time": "86.072",
"value": "Oklahoma State Highway 2"
},
{
"operation": "Highlighting",
"time": "98.454",
"value": "Oklahoma State Highway 2_sentence_14"
},
{
"operation": "Highlighting",
"time": "103.549",
"value": "Oklahoma State Highway 2_sentence_15"
},
{
"operation": "finish",
"time": "127.942",
"value": "finish"
}
]
}
|
6974
|
In Ludruk, a simple and straightforward comedy, actors would almost always use the Surabaya dialect of Javanese language (language of the Javanese people from the central and eastern parts of the island of Java).
|
{
"evidence": [
"Ludruk_sentence_5",
"Ludruk_sentence_4",
"Javanese language_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn Ludruk, a simple and straightforward comedy, actors would almost always use the Surabaya dialect of Javanese language (language of the Javanese people from the central and eastern parts of the island of Java).\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Ludruk_sentence_5",
"Ludruk_sentence_4",
"Javanese language_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 8190,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.05",
"value": "?search="
},
{
"operation": "search",
"time": "16.387",
"value": "Ludruk"
},
{
"operation": "Now on",
"time": "16.971",
"value": "Ludruk"
},
{
"operation": "Highlighting",
"time": "131.845",
"value": "Ludruk_sentence_3"
},
{
"operation": "Highlighting",
"time": "136.501",
"value": "Ludruk_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "137.158",
"value": "Ludruk_sentence_3"
},
{
"operation": "Highlighting",
"time": "140.798",
"value": "Ludruk_sentence_4"
},
{
"operation": "hyperlink",
"time": "149.398",
"value": "Javanese language"
},
{
"operation": "Now on",
"time": "151.907",
"value": "Javanese language"
},
{
"operation": "Highlighting",
"time": "157.902",
"value": "Javanese language_sentence_3"
},
{
"operation": "finish",
"time": "166.25",
"value": "finish"
}
]
}
|
74026
|
Arts editor for The Catholic Herald, Miguel Cullen, gave up writing to poetry focus on journalism.
|
{
"evidence": [
"Miguel Cullen_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nArts editor for The Catholic Herald, Miguel Cullen, gave up writing to poetry focus on journalism.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Miguel Cullen_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 8191,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.754",
"value": "?search="
},
{
"operation": "search",
"time": "16.825",
"value": "Miguel Cullen"
},
{
"operation": "Now on",
"time": "17.478",
"value": "Miguel Cullen"
},
{
"operation": "Highlighting",
"time": "43.451",
"value": "Miguel Cullen_sentence_0"
},
{
"operation": "finish",
"time": "58.651",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.