Datasets:

mariem123kfg
/

sob-ft-false-negatives

record_id string	context string	question string	json_schema string	validated_output string	source_split string	source_dataset string	question_type string	schema_complexity string	source_id string	errored_json string	error_count int64	error_difficulty string	error_type string	error_types string	error_location string	error_locations string	error_description string	error_descriptions string	error_difficulties string	error_source string	errors_json string	finetuning_task string	finetuning_question string	finetuning_input string	finetuning_target string	finetuning_messages string	repair_candidate string	true_has_error bool	false_negative_reason string	candidate_equals_gold bool
6202383b7795b58577d827665b964bf0bc93125b5e67cff9aeb6546e53497fd1	Donald Trump presidential campaign, 2000: Donald Trump's presidential campaign of 2000 for the nomination of the Reform Party began when real estate magnate Donald Trump of New York announced the creation of a presidential exploratory committee on the October 7, 1999 edition of "Larry King Live". Though Trump had neve...	Which of the high-profile candidates vied for the Reform Party presidential primaries, 2000 was a senior advisor to U.S. Presidents Richard Nixon?	{"type": "object", "properties": {"candidate_name": {"type": "string", "description": "The name of the candidate who vied for the Reform Party presidential primaries in 2000."}, "party_nomination_year": {"type": "integer", "description": "The year of the presidential nomination sought by the candidate."}, "party_name":...	{"candidate_name": "Pat Buchanan", "party_nomination_year": 2000, "party_name": "Reform Party", "advisor_role_details": {"role": "senior advisor", "presidents_advised": ["Richard Nixon", "Gerald Ford", "Ronald Reagan"]}}	train	hotpotqa	bridge	hard	5ac1d5a45542994d76dccefa	{"candidate_name": "Pat Buchanan", "party_nomination_year": 2000, "party_name": "Reform Party", "advisor_role_details": {"role": "senior advisor", "presidents_advised": ["Richard Nixon", "Gerald Ford", "Ronald Reagan"]}}	2	hard	semantic_negation_miss+semantic_scope_overgeneralization	["semantic_negation_miss", "semantic_scope_overgeneralization"]	"advisor_role_details.presidents_advised[1]"	["advisor_role_details.presidents_advised[1]", ["advisor_role_details", "presidents_advised"]]	The model incorrectly omitted Gerald Ford and Ronald Reagan from the list of presidents Pat Buchanan advised, failing to capture his full advisory history as described in the context. \| The extracted candidate 'Donald Trump' was never a senior advisor to U.S. Presidents Richard Nixon, Gerald Ford, or Ronald Reagan; thi...	["The model incorrectly omitted Gerald Ford and Ronald Reagan from the list of presidents Pat Buchanan advised, failing to capture his full advisory history as described in the context.", "The extracted candidate 'Donald Trump' was never a senior advisor to U.S. Presidents Richard Nixon, Gerald Ford, or Ronald Reagan; ...	["hard", "hard"]	llm	[{"error_type": "semantic_negation_miss", "error_difficulty": "hard", "error_location": "advisor_role_details.presidents_advised[1]", "error_description": "The model incorrectly omitted Gerald Ford and Ronald Reagan from the list of presidents Pat Buchanan advised, failing to capture his full advisory history as descri...	repair	Which of the high-profile candidates vied for the Reform Party presidential primaries, 2000 was a senior advisor to U.S. Presidents Richard Nixon?	CONTEXT: Donald Trump presidential campaign, 2000: Donald Trump's presidential campaign of 2000 for the nomination of the Reform Party began when real estate magnate Donald Trump of New York announced the creation of a presidential exploratory committee on the October 7, 1999 edition of "Larry King Live". Though Trump...	{"has_error": true, "location": "advisor_role_details.presidents_advised[1]", "corrected_json": {"candidate_name": "Pat Buchanan", "party_nomination_year": 2000, "party_name": "Reform Party", "advisor_role_details": {"role": "senior advisor", "presidents_advised": ["Richard Nixon", "Gerald Ford", "Ronald Reagan"]}}}	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	{"candidate_name": "Pat Buchanan", "party_nomination_year": 2000, "party_name": "Reform Party", "advisor_role_details": {"role": "senior advisor", "presidents_advised": ["Richard Nixon", "Gerald Ford", "Ronald Reagan"]}}	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
ft_extract_00042031	[![](/images/log/logoFen90.png)](/) ![](/images/log/logoFenNuevo.svg) [![](/fonts/en.svg)](/en) [![](/fonts/es.svg)](/es) [Universidad de Chile ](http://www.uchile.cl/) __ Toggle navigation * QUIÉNES SOMOS * * [Equipo Directivo](/es/quienes-somos/equipo-directivo) * [Plan Estratégico](/es/quienes-somos/plan-estrategico...	Listado de páginas web de ingenieroa comercial uchile	{"$defs":{"Dish":{"properties":{"name":{"description":"Titulo de la pagina","title":"Name","type":"string"},"url":{"description":"url de la pagina","title":"Url","type":"string"}},"required":["name","url"],"title":"Dish","type":"object"}},"properties":{"paginas":{"items":{"$ref":"#/$defs/Dish"},"title":"Paginas","type"...	{"paginas":[{"name":"Ingeniería Comercial","url":"http://www.admisionfen.cl/ingenieria-comercial/"}]}	train	scrapegraphai_full	null	medium	null	{"paginas": [{"name": "Ingeniería Comercial", "url": "http://www.admisionfen.cl/ingenieria-comercial/"}]}	3	medium	array_duplicate+string_truncation+array_drop_element	["array_duplicate", "string_truncation", "array_drop_element"]	["paginas"]	[["paginas"], ["paginas", 1, "url"], ["paginas"]]	Duplicated element [0] in array paginas (len 1 -> 2) \| Truncated paginas[1].url: 'http://www.admisionfen.cl/ingenieria-comercial/' -> 'http://www.admisionfen.' \| Dropped element [1] from array paginas (len 2 -> 1)	["Duplicated element [0] in array paginas (len 1 -> 2)", "Truncated paginas[1].url: 'http://www.admisionfen.cl/ingenieria-comercial/' -> 'http://www.admisionfen.'", "Dropped element [1] from array paginas (len 2 -> 1)"]	["medium", "medium", "medium"]	algorithmic	[{"error_type": "array_duplicate", "error_difficulty": "medium", "error_location": ["paginas"], "error_description": "Duplicated element [0] in array paginas (len 1 -> 2)", "error_source": "algorithmic"}, {"error_type": "string_truncation", "error_difficulty": "medium", "error_location": ["paginas", 1, "url"], "error_d...	repair	Listado de páginas web de ingenieroa comercial uchile	CONTEXT: [![](/images/log/logoFen90.png)](/) ![](/images/log/logoFenNuevo.svg) [![](/fonts/en.svg)](/en) [![](/fonts/es.svg)](/es) [Universidad de Chile ](http://www.uchile.cl/) __ Toggle navigation * QUIÉNES SOMOS * * [Equipo Directivo](/es/quienes-somos/equipo-directivo) * [Plan Estratégico](/es/quienes-somos/plan-es...	{"has_error": true, "location": ["paginas"], "corrected_json": {"paginas": [{"name": "Ingeniería Comercial", "url": "http://www.admisionfen.cl/ingenieria-comercial/"}]}}	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	{"paginas": [{"name": "Ingeniería Comercial", "url": "http://www.admisionfen.cl/ingenieria-comercial/"}]}	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
json-training-2976	null	As an aspiring electrician, I'm curious to know which vocational training programs offer specialized courses in smart home automation and renewable energy systems. Can you compile a list of institutions offering these courses, along with their respective locations and program lengths?	{"$id": "#", "type": "array", "items": {"$ref": "#/definitions/institution"}, "definitions": {"institution": {"type": "object", "properties": {"name": {"type": "string"}, "location": {"type": "string"}, "programLength": {"type": "integer"}}}}}	[ { "name": "Electrical Training Institute", "location": "New York, USA", "programLength": 12 }, { "name": "Smart Home Automation Academy", "location": "London, UK", "programLength": 18 } ]	train	ChristianAzinn/json-training	synthetic_ungrounded	easy	null	[{"name": "Electrical Training Institute", "location": "New York, USA", "programLength": 12}, {"name": "Smart Home Automation Academy", "location": "London, UK", "programLength": 18}]	2	hard	case_corruption+semantic_speaker_attribution_error	["case_corruption", "semantic_speaker_attribution_error"]	[0, "location"]	[[0, "location"], [0, "location"]]	Changed capitalization of [0].location: 'New York, USA' -> 'new york, usa' \| The locations of the two institutions were swapped, attributing 'Smart Home Automation Academy' to New York and 'Electrical Training Institute' to London, contradicting the context where each is correctly linked to its original city.	["Changed capitalization of [0].location: 'New York, USA' -> 'new york, usa'", "The locations of the two institutions were swapped, attributing 'Smart Home Automation Academy' to New York and 'Electrical Training Institute' to London, contradicting the context where each is correctly linked to its original city."]	["medium", "hard"]	mixed	[{"error_type": "case_corruption", "error_difficulty": "medium", "error_location": [0, "location"], "error_description": "Changed capitalization of [0].location: 'New York, USA' -> 'new york, usa'", "error_source": "algorithmic"}, {"error_type": "semantic_speaker_attribution_error", "error_difficulty": "hard", "error_l...	repair	As an aspiring electrician, I'm curious to know which vocational training programs offer specialized courses in smart home automation and renewable energy systems. Can you compile a list of institutions offering these courses, along with their respective locations and program lengths?	CONTEXT: QUESTION: As an aspiring electrician, I'm curious to know which vocational training programs offer specialized courses in smart home automation and renewable energy systems. Can you compile a list of institutions offering these courses, along with their respective locations and program lengths? JSON SCHEMA:...	{"has_error": true, "location": [0, "location"], "corrected_json": [{"name": "Electrical Training Institute", "location": "New York, USA", "programLength": 12}, {"name": "Smart Home Automation Academy", "location": "London, UK", "programLength": 18}]}	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	[{"name": "Electrical Training Institute", "location": "New York, USA", "programLength": 12}, {"name": "Smart Home Automation Academy", "location": "London, UK", "programLength": 18}]	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
ft_extract_00020153	# [![Smart Employee Benefits Inc.](/Portals/61/SEB- logo-v1.png?ver=2024-05-23-180646-087)](https://www.seb-inc.com/ "Smart Employee Benefits Inc.") [ Hidden Link](javascript:__doPostBack\$'dnn$dnnSearch$cmdSearch',''\$) __ Toggle Menu Button * [Hidden Link](http://www.behance.com "Behance") * [Hidden Link](https://w...	Extract the name of the company from the website header section, specifically from the logo area. Instructions: - Use only the provided content to extract the company name. - Look for elements that are commonly used for logos, such as <img> tags with ...	{"properties":{"website_title":{"default":null,"title":"Website Title","type":"string"}},"title":"WebsiteTitle","type":"object"}	{"website_title":"Smart Employee Benefits Inc."}	train	scrapegraphai_full	null	easy	null	{"website_title": "Smart Employee Benefits Inc."}	2	hard	word_typo+plausible_pool_substitution	["word_typo", "plausible_pool_substitution"]	["website_title"]	[["website_title"], ["website_title"]]	Introduced typo in website_title: 'Smart Employee Benefits Inc.' -> 'Smart Employee Benefits Inz.' \| Substituted plausible-but-wrong value for website_title: 'Smart Employee Benefits Inz.' -> 'Smart Employee Benefits Inc.'	["Introduced typo in website_title: 'Smart Employee Benefits Inc.' -> 'Smart Employee Benefits Inz.'", "Substituted plausible-but-wrong value for website_title: 'Smart Employee Benefits Inz.' -> 'Smart Employee Benefits Inc.'"]	["medium", "hard"]	algorithmic	[{"error_type": "word_typo", "error_difficulty": "medium", "error_location": ["website_title"], "error_description": "Introduced typo in website_title: 'Smart Employee Benefits Inc.' -> 'Smart Employee Benefits Inz.'", "error_source": "algorithmic"}, {"error_type": "plausible_pool_substitution", "error_difficulty": "ha...	repair	Extract the name of the company from the website header section, specifically from the logo area. Instructions: - Use only the provided content to extract the company name. - Look for elements that are commonly used for logos, such as <img> tags with attributes like "...	CONTEXT: # [![Smart Employee Benefits Inc.](/Portals/61/SEB- logo-v1.png?ver=2024-05-23-180646-087)](https://www.seb-inc.com/ "Smart Employee Benefits Inc.") [ Hidden Link](javascript:__doPostBack\$'dnn$dnnSearch$cmdSearch',''\$) __ Toggle Menu Button * [Hidden Link](http://www.behance.com "Behance") * [Hidden Link](...	{"has_error": true, "location": ["website_title"], "corrected_json": {"website_title": "Smart Employee Benefits Inc."}}	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	{"website_title": "Smart Employee Benefits Inc."}	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
ft_extract_00058430	o a podcast or look at a product description, how long you spent on this service and the web pages you visit etc. This is very helpful to understand the relevance of (non-advertising) content that is shown to you. View Illustrations Object to Legitimate Interests Remove Objection * ##### Understand audiences through st...	null	{"description":"Represents if present.","properties":{"is_present":{"description":"Whether is present or not.","enum":["yes","no"],"title":"Is Present","type":"string"}},"required":["is_present"],"title":"IsPresent","type":"object"}	{"is_present": "yes"}	train	scrapegraphai_finetuning_subset	null	easy	null	{"is_present": "yes"}	2	hard	enum_swap+plausible_pool_substitution	["enum_swap", "plausible_pool_substitution"]	["is_present"]	[["is_present"], ["is_present"]]	Swapped enum is_present to another allowed value: 'yes' -> 'no' \| Substituted plausible-but-wrong value for is_present: 'no' -> 'yes'	["Swapped enum is_present to another allowed value: 'yes' -> 'no'", "Substituted plausible-but-wrong value for is_present: 'no' -> 'yes'"]	["medium", "hard"]	algorithmic	[{"error_type": "enum_swap", "error_difficulty": "medium", "error_location": ["is_present"], "error_description": "Swapped enum is_present to another allowed value: 'yes' -> 'no'", "error_source": "algorithmic"}, {"error_type": "plausible_pool_substitution", "error_difficulty": "hard", "error_location": ["is_present"],...	repair	Extract structured data for IsPresent from the context.	CONTEXT: o a podcast or look at a product description, how long you spent on this service and the web pages you visit etc. This is very helpful to understand the relevance of (non-advertising) content that is shown to you. View Illustrations Object to Legitimate Interests Remove Objection * ##### Understand audiences t...	{"has_error": true, "location": ["is_present"], "corrected_json": {"is_present": "yes"}}	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	{"is_present": "yes"}	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
ft_extract_00047289	![](http://www.kdd.org/kdd2016/images/kdd2016_sf_blurred.jpg) Toggle navigation [KDD 2016](http://www.kdd.org/kdd2016) * [ATTENDING](http://www.kdd.org/kdd2016/registration) * [PROGRAM](http://www.kdd.org/kdd2016/program) * [CALLS](http://www.kdd.org/kdd2016/calls) * [NEWS](http://www.kdd.org/kdd2016/news) * [TUTOR...	KDD 2016 - San Francisco, CA, USA https://global.oup.com/	{"$defs":{"ScrapedChapter":{"description":"Schema of a chapter present in the table of contents.","properties":{"category":{"description":"Category of the chapter ('content' or 'other')","title":"Category","type":"string"},"prefix":{"description":"Prefix of the chapter (e.g. Part, Chapter, Chapter 2.3)","title":"Prefix...	{"certainty_percentage":100,"chapters":[{"category":"content","prefix":"Conference","title":"KDD 2016"},{"category":"content","prefix":"5","title":"Keynotes"},{"category":"content","prefix":"NEW for 2016","title":"Conversation Topics at KDD2016"},{"category":"content","prefix":"Track","title":"Applied Data Science Invi...	train	scrapegraphai_full	null	medium	null	{"certainty_percentage": 100, "chapters": [{"category": "content", "prefix": "Conference", "title": "KDD 2016"}, {"category": "content", "prefix": "5", "title": "Keynotes"}, {"category": "content", "prefix": "NEW for 2016", "title": "Conversation Topics at KDD2016"}, {"category": "content", "prefix": "Track", "title": ...	2	hard	type_break+semantic_role_title_inflation	["type_break", "semantic_role_title_inflation"]	["chapters", 3, "prefix"]	[["chapters", 3, "prefix"], ["chapters", 3, "prefix"]]	Replaced chapters[3].prefix (string -> number) with implausible value 9999 \| The role 'Invited Talks' was inflated to 'Invited Panels', incorrectly suggesting a broader discussion format than the context specifies for the Applied Data Science track.	["Replaced chapters[3].prefix (string -> number) with implausible value 9999", "The role 'Invited Talks' was inflated to 'Invited Panels', incorrectly suggesting a broader discussion format than the context specifies for the Applied Data Science track."]	["easy", "hard"]	mixed	[{"error_type": "type_break", "error_difficulty": "easy", "error_location": ["chapters", 3, "prefix"], "error_description": "Replaced chapters[3].prefix (string -> number) with implausible value 9999", "error_source": "algorithmic"}, {"error_type": "semantic_role_title_inflation", "error_difficulty": "hard", "error_loc...	repair	KDD 2016 - San Francisco, CA, USA https://global.oup.com/	CONTEXT: ![](http://www.kdd.org/kdd2016/images/kdd2016_sf_blurred.jpg) Toggle navigation [KDD 2016](http://www.kdd.org/kdd2016) * [ATTENDING](http://www.kdd.org/kdd2016/registration) * [PROGRAM](http://www.kdd.org/kdd2016/program) * [CALLS](http://www.kdd.org/kdd2016/calls) * [NEWS](http://www.kdd.org/kdd2016/news)...	{"has_error": true, "location": ["chapters", 3, "prefix"], "corrected_json": {"certainty_percentage": 100, "chapters": [{"category": "content", "prefix": "Conference", "title": "KDD 2016"}, {"category": "content", "prefix": "5", "title": "Keynotes"}, {"category": "content", "prefix": "NEW for 2016", "title": "Conversat...	[{"role": "system", "content": "You verify and repair JSON answers. Given a CONTEXT, QUESTION, JSON SCHEMA, and a candidate JSON, decide whether the candidate JSON is fully correct and grounded in the context. If it is wrong, return the corrected JSON. If it is already correct, return it unchanged and report no error. ...	{"certainty_percentage": 100, "chapters": [{"category": "content", "prefix": "Conference", "title": "KDD 2016"}, {"category": "content", "prefix": "5", "title": "Keynotes"}, {"category": "content", "prefix": "NEW for 2016", "title": "Conversation Topics at KDD2016"}, {"category": "content", "prefix": "Track", "title": ...	true	candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetuning_target.has_error=True	true
stage-eval-data1	"2023년 한일 정부 초청 장학생 1년 학부 과정 프로그램 중 한국해양대학교에(...TRUNCATED)	"2023년 한국해양대학교 한일 정부 초청 장학생 1년 학부 과정 전공 분야별 (...TRUNCATED)	"{\"$schema\": \"https://json-schema.org/draft/2020-12/schema\", \"type\": \"object\", \"properties\(...TRUNCATED)	"{\n \"프로그램\": {\n \"National Korea Maritime & Ocean University\": {\n \"Ocean Scie(...TRUNCATED)	train	boradorish/STAGE-eval	train	hard	data1	"{\"프로그램\": {\"National Korea Maritime & Ocean University\": {\"Ocean Science\": {\"Naval Ar(...TRUNCATED)	2	medium	type_break+enum_swap	["type_break", "enum_swap"]	"[\"\\ud504\\ub85c\\uadf8\\ub7a8\", \"National Korea Maritime & Ocean University\", \"Engineering Sc(...TRUNCATED)	"[[\"\\ud504\\ub85c\\uadf8\\ub7a8\", \"National Korea Maritime & Ocean University\", \"Engineering S(...TRUNCATED)	"Replaced 프로그램.National Korea Maritime & Ocean University.Engineering Science.Logistics (str(...TRUNCATED)	"[\"Replaced \\ud504\\ub85c\\uadf8\\ub7a8.National Korea Maritime & Ocean University.Engineering Sci(...TRUNCATED)	["easy", "medium"]	algorithmic	"[{\"error_type\": \"type_break\", \"error_difficulty\": \"easy\", \"error_location\": [\"프로그(...TRUNCATED)	repair	"2023년 한국해양대학교 한일 정부 초청 장학생 1년 학부 과정 전공 분야별 (...TRUNCATED)	"CONTEXT:\n2023년 한일 정부 초청 장학생 1년 학부 과정 프로그램 중 한국해양(...TRUNCATED)	"{\"has_error\": true, \"location\": [\"프로그램\", \"National Korea Maritime & Ocean University(...TRUNCATED)	"[{\"role\": \"system\", \"content\": \"You verify and repair JSON answers. Given a CONTEXT, QUESTIO(...TRUNCATED)	"{\"프로그램\": {\"National Korea Maritime & Ocean University\": {\"Ocean Science\": {\"Naval Ar(...TRUNCATED)	true	"candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetu(...TRUNCATED)	true
ft_extract_00060325	"ini/providers)\| [ Model](/models/jamba-1-5-mini)[ API Providers](/models/jamba-1-5-mini/providers) (...TRUNCATED)	null	"{\"properties\":{\"QueryAnswer\":{\"description\":\"The answer to the query\",\"title\":\"Queryansw(...TRUNCATED)	"{\"QueryAnswer\": \"Summary of the content: a catalog/listing of API providers and models, includin(...TRUNCATED)	train	scrapegraphai_finetuning_subset	null	easy	null	"{\"QueryAnswer\": \"Summary of the content: a catalog/listing of API providers and models, includin(...TRUNCATED)	3	hard	word_typo+case_corruption+plausible_pool_substitution	["word_typo", "case_corruption", "plausible_pool_substitution"]	["QueryAnswer"]	[["QueryAnswer"], ["QueryAnswer"], ["QueryAnswer"]]	"Introduced typo in QueryAnswer: 'Summary of the content: a catalog/listing of API providers and mod(...TRUNCATED)	"[\"Introduced typo in QueryAnswer: 'Summary of the content: a catalog/listing of API providers and (...TRUNCATED)	["medium", "medium", "hard"]	algorithmic	"[{\"error_type\": \"word_typo\", \"error_difficulty\": \"medium\", \"error_location\": [\"QueryAnsw(...TRUNCATED)	repair	Extract structured data for ResponseModel from the context.	"CONTEXT:\nini/providers)\| [ Model](/models/jamba-1-5-mini)[ API Providers](/models/jamba-1-5-mini/p(...TRUNCATED)	"{\"has_error\": true, \"location\": [\"QueryAnswer\"], \"corrected_json\": {\"QueryAnswer\": \"Summ(...TRUNCATED)	"[{\"role\": \"system\", \"content\": \"You verify and repair JSON answers. Given a CONTEXT, QUESTIO(...TRUNCATED)	"{\"QueryAnswer\": \"Summary of the content: a catalog/listing of API providers and models, includin(...TRUNCATED)	true	"candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetu(...TRUNCATED)	true
stage-eval-data1554	"## 전립선 비대증 치료제 시장 동향 분석\n\n최근 전립선 비대증 치료제 시(...TRUNCATED)	"전립선 비대증 치료제 시장 동향 및 삼성실로도신캡슐4밀리그램의 급여 상(...TRUNCATED)	"{\"$schema\": \"https://json-schema.org/draft/2020-12/schema\", \"type\": \"object\", \"properties\(...TRUNCATED)	"{\n \"약제_급여_목록_및_급여_상한금액표_중_신설\": {\n \"주성분코드\": \"(...TRUNCATED)	train	boradorish/STAGE-eval	train	medium	data1554	"{\"약제_급여_목록_및_급여_상한금액표_중_신설\": {\"주성분코드\": \"제품코(...TRUNCATED)	2	medium	word_typo+enum_swap	["word_typo", "enum_swap"]	"[\"\\uc57d\\uc81c_\\uae09\\uc5ec_\\ubaa9\\ub85d_\\ubc0f_\\uae09\\uc5ec_\\uc0c1\\ud55c\\uae08\\uc561(...TRUNCATED)	"[[\"\\uc57d\\uc81c_\\uae09\\uc5ec_\\ubaa9\\ub85d_\\ubc0f_\\uae09\\uc5ec_\\uc0c1\\ud55c\\uae08\\uc56(...TRUNCATED)	"Introduced typo in 약제_급여_목록_및_급여_상한금액표_중_신설.상한금액: '퇴장(...TRUNCATED)	"[\"Introduced typo in \\uc57d\\uc81c_\\uae09\\uc5ec_\\ubaa9\\ub85d_\\ubc0f_\\uae09\\uc5ec_\\uc0c1\\(...TRUNCATED)	["medium", "medium"]	algorithmic	"[{\"error_type\": \"word_typo\", \"error_difficulty\": \"medium\", \"error_location\": [\"약제_(...TRUNCATED)	repair	"전립선 비대증 치료제 시장 동향 및 삼성실로도신캡슐4밀리그램의 급여 상(...TRUNCATED)	"CONTEXT:\n## 전립선 비대증 치료제 시장 동향 분석\n\n최근 전립선 비대증 치(...TRUNCATED)	"{\"has_error\": true, \"location\": [\"약제_급여_목록_및_급여_상한금액표_중_신설\"(...TRUNCATED)	"[{\"role\": \"system\", \"content\": \"You verify and repair JSON answers. Given a CONTEXT, QUESTIO(...TRUNCATED)	"{\"약제_급여_목록_및_급여_상한금액표_중_신설\": {\"주성분코드\": \"제품코(...TRUNCATED)	true	"candidate JSON equals validated_output (no actual error present) but true_has_error=True and finetu(...TRUNCATED)	true

SOB FT False Negatives

Subset of mariem123kfg/sob-ft-finetune-ready where the candidate JSON has no actual error, but labels claim an error.

Definition used

A row is included when all of the following hold:

true_has_error == True
repair_candidate (or errored_json) equals validated_output (canonical JSON compare)
finetuning_target.has_error == True and corrected_json equals the already-correct candidate

These look like failed / cancelled-out error injections: metadata still describes injected errors, but the final candidate is identical to gold.

Counts

Source dataset size: 97,426 rows (50,552 extract + 46,874 repair)
Extract rows with true_has_error=True: 0
Repair rows with true_has_error=True: 46,874
False negatives matching the definition above: 9

Split

repair_false_negatives: the 9 mismatched repair rows (full original columns + annotation fields)

Downloads last month: 13

Number of rows:

Total file size:

167 kB