Datasets:

mjbommar
/

opengloss-instruct

id stringlengths 22 36	family stringclasses 22 values	task stringclasses 7 values	source stringclasses 4 values	messages listlengths 2 7	text stringlengths 67 9.23k	seeds listlengths 1 6	generator stringclasses 1 value	n_tokens_alea int32 17 2k	in_r2_decay bool 2 classes
og_gqa-e6a4f21b12db54c1	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "From what the passage says, how can flat figures be examined in mathematical work? Keep it brief.\n<\|context\|>\n<\|snippet\|>[1] flat figure\nToday, the term flat figure anchors rigorous vocabulary in educational contexts. The concept involves comparing shapes, computing area and...	<\|start\|><\|user\|> From what the passage says, how can flat figures be examined in mathematical work? Keep it brief. <\|context\|> <\|snippet\|>[1] flat figure Today, the term flat figure anchors rigorous vocabulary in educational contexts. The concept involves comparing shapes, computing area and perimeter, and exploring t...	[ "og-qa:flat_figure:noun:0#qa4", "og-qa:flat_figure:adjective:1#qa3", "og-enc:flat_figure\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	259	false
og_gqa-7b136a0e8b47bc8c	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "You help people read documents closely." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] faulty data\nData may be deemed faulty when its generation, recording, transmission, storage, or preprocessing introduces defects that violate expected constraint...	<\|start\|><\|system\|> You help people read documents closely.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] faulty data Data may be deemed faulty when its generation, recording, transmission, storage, or preprocessing introduces defects that violate expected constraints, formats, or known properties of the underlying system...	[ "og-qa:faulty_data:noun:1#qa5", "og-enc:faulty_data\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	326	false
wd_extract-de54f6d81cb0a900	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Esplanade Zagreb Hotel\nEsplanade Zagreb Hotel is a hotel Esplanade in Zagreb. Esplanade Zagreb Hotel dates from April 22, 1925. Esplanade Zagreb Hotel is part of Donji grad. Esplanade Zagreb Hotel opened on April 22, 1925. Esplanade Zagreb Hotel is ...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Esplanade Zagreb Hotel Esplanade Zagreb Hotel is a hotel Esplanade in Zagreb. Esplanade Zagreb Hotel dates from April 22, 1925. Esplanade Zagreb Hotel is part of Donji grad. Esplanade Zagreb Hotel opened on April 22, 1925. Esplanade Zagreb Hotel is in Croatia.<\|/snippet\|> <\|...	[ "wd:Q3739460" ]	deterministic (opengloss-instruct v1)	225	true
og_gqa-a0474ac725b0ac5e	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "Looking at the text, why can heating help towels avoid developing a stale smell?\n<\|context\|>\n<\|snippet\|>[1] towel rail\nSome models are freestanding or attached to the side of a cabinet, making them useful where drilling is not possible.\n\nA heated towel rail, often called a...	<\|start\|><\|user\|> Looking at the text, why can heating help towels avoid developing a stale smell? <\|context\|> <\|snippet\|>[1] towel rail Some models are freestanding or attached to the side of a cabinet, making them useful where drilling is not possible. A heated towel rail, often called a heated towel rack, combines ...	[ "og-qa:towel_rail:noun:0#qa5", "og-qa:towel_rail:noun:0#qa6", "og-enc:towel_rail\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	382	true
og_gqa-2bf412c4159b1326	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] locust\nLocust refers to several species of short-horned grasshoppers in the family Acrididae that, at high population density, undergo dramatic changes in behavior, physiology, and morphology. This density-dependent transformation enables the insect...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] locust Locust refers to several species of short-horned grasshoppers in the family Acrididae that, at high population density, undergo dramatic changes in behavior, physiology, and morphology. This density-dependent transformation enables the insects to form massive, coordin...	[ "og-qa:locust:noun:0#qa2", "og-enc:locust\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	402	false
og_rag_abstain-845b285ec62567db	og_rag_abstain	abstention	opengloss	[ { "role": "system", "content": "Be faithful to the text." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] pupped\nPupped is a rare word about an animal having young pups. It tells us pups were born. Animal notes may use it. Formal writing often says whelped or had a litter instead. ...	<\|start\|><\|system\|> Be faithful to the text.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] pupped Pupped is a rare word about an animal having young pups. It tells us pups were born. Animal notes may use it. Formal writing often says whelped or had a litter instead. It talks about the young, not the mother's age or health...	[ "og-qa:dino:noun:0#qa12", "og-enc:pupped\|grade_1\|plain", "og-enc:feral_cat\|college\|plain" ]	deterministic (opengloss-instruct v1)	332	true
og_gqa-e44dcfe47ba71f7f	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] belly dance\nCostumes used in professional stage performances often include a fitted top, a skirt or harem-style trousers, and a hip scarf or belt, while social dancers may wear ordinary clothing.\n\nModern belly dance developed through the interacti...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] belly dance Costumes used in professional stage performances often include a fitted top, a skirt or harem-style trousers, and a hip scarf or belt, while social dancers may wear ordinary clothing. Modern belly dance developed through the interaction of older regional dance p...	[ "og-qa:belly_dance:verb:0#qa5", "og-enc:belly_dance\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	256	true
og_gqa-5847ab313e4d1739	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "Going by this text, why can a density function be used to calculate expectations? Keep your answer grounded in the reading.\n<\|context\|>\n<\|snippet\|>[1] densities\nKey aspects and interpretations include:\nPhysical densities: variables such as mass density ρ and number density ...	<\|start\|><\|user\|> Going by this text, why can a density function be used to calculate expectations? Keep your answer grounded in the reading. <\|context\|> <\|snippet\|>[1] densities Key aspects and interpretations include: Physical densities: variables such as mass density ρ and number density n quantify substance concent...	[ "og-qa:densities:noun:2#qa3", "og-enc:densities\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	342	false
og_gqa-399221c14b0d814a	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] confuser\nBy foregrounding the sources and mechanics of confusion, the notion of a confuser supports rigorous evaluation of robustness, clarity, and resilience across research and practice.\n\nIn practice, a confuser can take multiple forms and serve...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] confuser By foregrounding the sources and mechanics of confusion, the notion of a confuser supports rigorous evaluation of robustness, clarity, and resilience across research and practice. In practice, a confuser can take multiple forms and serve diverse roles. Key aspects ...	[ "og-qa:confuser:noun:0#qa3", "og-qa:confuser:noun:0#qa4", "og-enc:confuser\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	434	false
kl3m_rag_define-fd29d56878d4f13d	kl3m_rag_define	citation	kl3m	[ { "role": "user", "content": "What do the documents say \"Legal proceedings\" means? Give the source number.\n<\|context\|>\n<\|snippet\|>[1] § 417.9\n#### § 417.9 Launch site responsibility.\n\n(a) A launch operator must ensure that launch processing at a launch site in the United States satisfies the requirem...	<\|start\|><\|user\|> What do the documents say "Legal proceedings" means? Give the source number. <\|context\|> <\|snippet\|>[1] § 417.9 #### § 417.9 Launch site responsibility. (a) A launch operator must ensure that launch processing at a launch site in the United States satisfies the requirements of this part. Launch proce...	[ "kl3m:ecfrx:903e99bb5a55d251", "kl3m:ecfrx:e537877549b95698", "kl3m:ecfrx:1b0119183cab3885" ]	deterministic (opengloss-instruct v1)	373	true
og_abstain-53538a41d7eb0cd1	og_abstain	abstention	opengloss	[ { "role": "user", "content": "One question about this: If a weather vane were placed on a roof rather than in obstructed air, what would be true of its reading?\n<\|context\|>\n<\|snippet\|>[1] flathead\nFlathead joins the words flat and head. It describes things, animals, or people with broad, flattened ends o...	<\|start\|><\|user\|> One question about this: If a weather vane were placed on a roof rather than in obstructed air, what would be true of its reading? <\|context\|> <\|snippet\|>[1] flathead Flathead joins the words flat and head. It describes things, animals, or people with broad, flattened ends or heads. The term often na...	[ "og-qa:weather_vane:noun:0#qa6", "og-enc:flathead\|grade_5\|plain" ]	deterministic (opengloss-instruct v1)	333	true
og_gqa-88834310b332b828	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "Answer based on the source: If the Federal Reserve’s decisions affect borrowing costs and the FOMC sets monetary policy, what can follow when the FOMC changes its policy stance?\n<\|context\|>\n<\|snippet\|>[1] Federal Reserve\nIts decisions affect borrowing costs for households an...	<\|start\|><\|user\|> Answer based on the source: If the Federal Reserve’s decisions affect borrowing costs and the FOMC sets monetary policy, what can follow when the FOMC changes its policy stance? <\|context\|> <\|snippet\|>[1] Federal Reserve Its decisions affect borrowing costs for households and businesses, the value of ...	[ "og-qa:federal_reserve:noun:0#qa2", "og-qa:federal_reserve:noun:0#qa6", "og-enc:federal_reserve\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	282	false
og_gqa-3cfd9022a8f2014a	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "Be brief and factual." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] clue\nIn everyday use, a clue is a piece of information that helps solve a problem, answer a question, or understand a situation. Clue supports inquiry and reading by pointing read...	<\|start\|><\|system\|> Be brief and factual.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] clue In everyday use, a clue is a piece of information that helps solve a problem, answer a question, or understand a situation. Clue supports inquiry and reading by pointing readers toward what remains unknown and guiding them toward ...	[ "og-qa:clue:verb:1#qa11", "og-enc:clue\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	303	false
og_gqa-d96e5bb4f9974394	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "You are an editor's assistant. Be exact." }, { "role": "user", "content": "Here is a question for you. If a cartridge has been placed in the chamber but the breech closer has not been returned and locked, what function has not yet been completed? Explain briefly.\...	<\|start\|><\|system\|> You are an editor's assistant. Be exact.<\|eot\|> <\|user\|> Here is a question for you. If a cartridge has been placed in the chamber but the breech closer has not been returned and locked, what function has not yet been completed? Explain briefly. <\|context\|> <\|snippet\|>[1] breech closer A breech clos...	[ "og-qa:breech_closer:noun:0#qa5", "og-enc:breech_closer\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	435	true
og_rag-e0762568d777f3f5	og_rag	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] tamable\nTamable describes a living organism that can be made more accepting of human contact under certain conditions. This ability, called tamability, varies among members of a species and reflects their development, social surroundings, and inheri...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] tamable Tamable describes a living organism that can be made more accepting of human contact under certain conditions. This ability, called tamability, varies among members of a species and reflects their development, social surroundings, and inherited traits. The word is al...	[ "og-qa:cottonseed:noun:1#qa0", "og-enc:tamable\|grade_10\|plain", "og-enc:anastatica\|grade_10\|plain", "og-enc:cottonseed\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	494	false
og_gqa-ebf431768d1db28a	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] time period\nTemporal scope and locale: considerations include time zones, daylight saving adjustments, and calendar systems, which affect boundary placement and comparability.\nBoundary handling and integrity: rules for boundary inclusivity, leap da...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] time period Temporal scope and locale: considerations include time zones, daylight saving adjustments, and calendar systems, which affect boundary placement and comparability. Boundary handling and integrity: rules for boundary inclusivity, leap days, and incomplete data wit...	[ "og-qa:time_period:noun:0#qa6", "og-enc:time_period\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	242	false
og_rag_abstain-2cecbc251d6915df	og_rag_abstain	abstention	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] The King in Yellow\nThe King in Yellow is a 1895 collection of strange fiction by American author Robert W. Chambers. The opening stories refer to a forbidden play that shares the book’s title. In the stories, reading it can expose people to frighten...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] The King in Yellow The King in Yellow is a 1895 collection of strange fiction by American author Robert W. Chambers. The opening stories refer to a forbidden play that shares the book’s title. In the stories, reading it can expose people to frightening truths, destroy their ...	[ "og-qa:high_culture:adjective:0#qa5", "og-enc:the_king_in_yellow\|grade_10\|plain", "og-enc:philosophise\|grade_5\|plain", "og-enc:inherent_property\|grade_5\|plain" ]	deterministic (opengloss-instruct v1)	422	true
wd_extract-24fa4c2138532e72	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Lunar and Planetary Institute\nLunar and Planetary Institute: American research institute in Houston. Lunar and Planetary Institute is part of Houston. Lunar and Planetary Institute is in United States. Lunar and Planetary Institute was established i...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Lunar and Planetary Institute Lunar and Planetary Institute: American research institute in Houston. Lunar and Planetary Institute is part of Houston. Lunar and Planetary Institute is in United States. Lunar and Planetary Institute was established in 1978.<\|/snippet\|> <\|/con...	[ "wd:Q239221" ]	deterministic (opengloss-instruct v1)	136	true
og_rag_abstain-994bec07a4b71ec1	og_rag_abstain	abstention	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] European Law Enforcement Organisation\nThe European Law Enforcement Organisation, usually called Europol, is the European Union agency that helps law enforcement bodies cooperate. Based in The Hague, it helps national police and other authorities tac...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] European Law Enforcement Organisation The European Law Enforcement Organisation, usually called Europol, is the European Union agency that helps law enforcement bodies cooperate. Based in The Hague, it helps national police and other authorities tackle serious crime that cro...	[ "og-qa:atrocity:noun:0#qa3", "og-enc:european_law_enforcement_organisation\|grade_10\|plain", "og-enc:tariff_schedules\|neutral\|plain", "og-enc:foreign_intelligence_surveillance_court\|grade_10\|plain" ]	deterministic (opengloss-instruct v1)	408	false
og_gqa-2567e71c77a3334c	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "Read the passage and answer. What can neighborhood organization patterns support through local collective action?\n<\|context\|>\n<\|snippet\|>[1] neighborhood organization\nSafety, services, and mutual aid: activities such as neighborhood watch, street-cleanup teams, and volunteer...	<\|start\|><\|user\|> Read the passage and answer. What can neighborhood organization patterns support through local collective action? <\|context\|> <\|snippet\|>[1] neighborhood organization Safety, services, and mutual aid: activities such as neighborhood watch, street-cleanup teams, and volunteer networks that support vuln...	[ "og-qa:neighborhood_organization:adjective:0#qa5", "og-enc:neighborhood_organization\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	304	true
og_rag-2c5510cc9bbd9e09	og_rag	citation	opengloss	[ { "role": "user", "content": "Reading these documents, what can excessive growth hormone-releasing factor signaling lead to?\n<\|context\|>\n<\|snippet\|>[1] automysophobia\nAutomysophobia is a very strong, lasting fear of feeling dirty. People may fear dust, dirt, spit, germs, or dirty things. They may think t...	<\|start\|><\|user\|> Reading these documents, what can excessive growth hormone-releasing factor signaling lead to? <\|context\|> <\|snippet\|>[1] automysophobia Automysophobia is a very strong, lasting fear of feeling dirty. People may fear dust, dirt, spit, germs, or dirty things. They may think touching these things made t...	[ "og-qa:growth_hormone_releasing_factor:noun:0#qa5", "og-enc:automysophobia\|grade_1\|plain", "og-enc:skullcap\|grade_10\|plain", "og-enc:growth_hormone_releasing_factor\|neutral\|plain", "og-enc:calmness\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	552	true
og_rag-80ae434963e711c4	og_rag	citation	opengloss	[ { "role": "system", "content": "You are a knowledgeable assistant." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] staunton\nKey features and applications:\nDistinct silhouettes that retain legibility at speed, enabling quick piece identification during play.\nProportional harmony ...	<\|start\|><\|system\|> You are a knowledgeable assistant.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] staunton Key features and applications: Distinct silhouettes that retain legibility at speed, enabling quick piece identification during play. Proportional harmony among piece classes (king, queen, bishop, knight, rook, pa...	[ "og-qa:staunton:adjective:0#qa2", "og-enc:staunton\|neutral\|plain", "og-enc:hammerlock\|grade_10\|plain", "og-enc:jump_shot\|grade_1\|plain", "og-enc:ring_a_rosy\|college\|plain", "og-enc:hakeem_olajuwon\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	953	false
og_gqa-07b53d8353d1bdd9	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] owain\nAnother central figure is Owain Glyndŵr (c. 1359–c. 1415), leader of the early fifteenth-century Welsh revolt against English rule and a lasting symbol of Welsh political identity and resistance. Through such associations, the name Owain acqui...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] owain Another central figure is Owain Glyndŵr (c. 1359–c. 1415), leader of the early fifteenth-century Welsh revolt against English rule and a lasting symbol of Welsh political identity and resistance. Through such associations, the name Owain acquired strong connotations of...	[ "og-qa:owain:noun:1#qa5", "og-qa:owain:noun:0#qa8", "og-qa:owain:noun:0#qa5", "og-enc:owain\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	335	false
wd_extract-734eddb44f0852da	wd_extract	extraction	wikidata	[ { "role": "user", "content": "Make a YAML record of name, mouthOfTheWatercourse, country, timeOfDiscovery based on the passage. Use null for missing data.\n<\|context\|>\n<\|snippet\|>[1] Swilly\nSwilly is a river in Ireland. Swilly is part of County Donegal. Swilly flows into Atlantic Ocean. Swilly lies in Ire...	<\|start\|><\|user\|> Make a YAML record of name, mouthOfTheWatercourse, country, timeOfDiscovery based on the passage. Use null for missing data. <\|context\|> <\|snippet\|>[1] Swilly Swilly is a river in Ireland. Swilly is part of County Donegal. Swilly flows into Atlantic Ocean. Swilly lies in Ireland.<\|/snippet\|> <\|/contex...	[ "wd:Q2155592" ]	deterministic (opengloss-instruct v1)	136	true
og_gqa-df04d5c0f27e6ba5	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] evidence base\nKey components of an evidence base include:\nTypes of evidence: randomized controlled trials, quasi-experimental studies, systematic reviews, meta-analyses, longitudinal studies, case studies.\nEvidence quality indicators: internal val...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] evidence base Key components of an evidence base include: Types of evidence: randomized controlled trials, quasi-experimental studies, systematic reviews, meta-analyses, longitudinal studies, case studies. Evidence quality indicators: internal validity, external validity, re...	[ "og-qa:evidence_base:noun:0#qa1", "og-enc:evidence_base\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	302	true
wd_extract-52421a37b18fd08c	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Flatanger Municipality\nFlatanger Municipality is a municipality in Trøndelag, Norway. Flatanger Municipality dates from January 1, 1871. Flatanger Municipality is in Norway. The capital of Flatanger Municipality is Lauvsnes.<\|/snippet\|>\n<\|/context\|...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Flatanger Municipality Flatanger Municipality is a municipality in Trøndelag, Norway. Flatanger Municipality dates from January 1, 1871. Flatanger Municipality is in Norway. The capital of Flatanger Municipality is Lauvsnes.<\|/snippet\|> <\|/context\|> Output a JSON mapping wit...	[ "wd:Q48276" ]	deterministic (opengloss-instruct v1)	159	false
og_gqa-a48d0c22c76bf7a5	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "Keep answers short and grounded in the text you are given." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] rearm\nIn security studies, the noun rearmament refers to deliberate policy efforts to reconstruct or expand a state's armed forces and defense...	<\|start\|><\|system\|> Keep answers short and grounded in the text you are given.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] rearm In security studies, the noun rearmament refers to deliberate policy efforts to reconstruct or expand a state's armed forces and defense industries after restraint, sanctions, or treaty limits...	[ "og-qa:rearm:verb:0#qa5", "og-enc:rearm\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	300	true
og_gqa-8e578b844266a11b	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "Stick to what the sources say." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] simple absence\nThe person may suddenly stop speaking or moving, stare blankly, and fail to respond. They often resume activity as if nothing happened and may not realize ...	<\|start\|><\|system\|> Stick to what the sources say.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] simple absence The person may suddenly stop speaking or moving, stare blankly, and fail to respond. They often resume activity as if nothing happened and may not realize that anything unusual occurred. Rapid blinking, small mo...	[ "og-qa:simple_absence:noun:0#qa6", "og-qa:simple_absence:noun:0#qa2", "og-enc:simple_absence\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	351	false
og_rag2-cffd409678ec395c	og_rag2	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Capetian dynasty\nHugh Capet was chosen by leading nobles after the Carolingian line failed to provide a surviving male heir. Early Capetian kings strengthened their position by arranging for their eldest sons to be crowned during their own lifetimes...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Capetian dynasty Hugh Capet was chosen by leading nobles after the Carolingian line failed to provide a surviving male heir. Early Capetian kings strengthened their position by arranging for their eldest sons to be crowned during their own lifetimes, helping to make successi...	[ "og-qa:valid_reasoning:noun:0#qa4", "og-qa:capetian_dynasty:noun:0#qa5", "og-enc:capetian_dynasty\|neutral\|plain", "og-enc:valid_reasoning\|neutral\|plain", "og-enc:lech_czech_and_rus\|grade_5\|plain", "og-enc:potboiler\|grade_5\|plain" ]	deterministic (opengloss-instruct v1)	598	false
og_gqa-8a9aec98e9c762b4	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] legal term\nLegal term\n\nA legal term denotes a word or phrase with a precise, legally defined meaning within a given jurisdiction. It underpins statutory drafting, contract interpretation, and judicial decision-making by delimiting rights, duties, ...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] legal term Legal term A legal term denotes a word or phrase with a precise, legally defined meaning within a given jurisdiction. It underpins statutory drafting, contract interpretation, and judicial decision-making by delimiting rights, duties, and remedies. Because legal ...	[ "og-qa:legal_term:noun:0#qa2", "og-enc:legal_term\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	307	true
og_rag-200ad0ee4fb4b775	og_rag	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] hugger\nThe term also appears in metaphorical uses and idioms, most notably in the compound tree hugger, which labels environmental advocacy within popular discourse and can carry evaluative or political overtones.\n\nSemantic scope: the word designa...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] hugger The term also appears in metaphorical uses and idioms, most notably in the compound tree hugger, which labels environmental advocacy within popular discourse and can carry evaluative or political overtones. Semantic scope: the word designates both the literal actor (...	[ "og-qa:hugger:noun:0#qa3", "og-enc:hugger\|neutral\|plain", "og-enc:roue\|college\|plain", "og-enc:icelander\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	533	false
og_rag-80cc3c51ec9e9b43	og_rag	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Jesus Prayer\nThe Jesus Prayer is a brief Christian appeal to Jesus Christ. Its best-known English wording is “Lord Jesus Christ, Son of God, have mercy on me, a sinner,” though many worshippers use a shorter version: “Lord Jesus Christ, have mercy o...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Jesus Prayer The Jesus Prayer is a brief Christian appeal to Jesus Christ. Its best-known English wording is “Lord Jesus Christ, Son of God, have mercy on me, a sinner,” though many worshippers use a shorter version: “Lord Jesus Christ, have mercy on me.” It is most closely ...	[ "og-qa:metaphysics:noun:1#qa3", "og-enc:jesus_prayer\|grade_10\|plain", "og-enc:metaphysics\|neutral\|plain", "og-enc:mantra\|grade_1\|plain" ]	deterministic (opengloss-instruct v1)	581	true
kl3m_which_section-e1f9a375f59c9653	kl3m_which_section	citation	kl3m	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] § 244.13\n(a) Temporary treatment benefits terminate upon a final determination with respect to the alien's eligibility for Temporary Protected Status.\n\n(b) Temporary treatment benefits terminate, in any case, sixty (60) days after the date that no...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] § 244.13 (a) Temporary treatment benefits terminate upon a final determination with respect to the alien's eligibility for Temporary Protected Status. (b) Temporary treatment benefits terminate, in any case, sixty (60) days after the date that notice is published of the ter...	[ "kl3m:ecfrx:71a97c05e6848e60", "kl3m:ecfr:62dae11f6e69beba", "kl3m:ecfrx:06a77b0e66d2a67d" ]	deterministic (opengloss-instruct v1)	394	true
og_gqa-becf4f25300bc4ee	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] degraded reefs\nCoral reefs are built by tiny organisms called polyps that secrete calcium carbonate, forming the hard framework of the reef; when stress from warmer water or pollution increases, corals may expel their symbiotic algae, a process know...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] degraded reefs Coral reefs are built by tiny organisms called polyps that secrete calcium carbonate, forming the hard framework of the reef; when stress from warmer water or pollution increases, corals may expel their symbiotic algae, a process known as coral bleaching. If s...	[ "og-qa:degraded_reefs:noun:0#qa2", "og-qa:degraded_reefs:noun:1#qa4", "og-qa:degraded_reefs:noun:1#qa3", "og-enc:degraded_reefs\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	529	false
kl3m_which_section-d133a73cdb9d5312	kl3m_which_section	citation	kl3m	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] 38 CFR 12.4\n(a) Upon authorization by the facility head or his or her designated representative, all funds, as defined in § 12.0 (except funds deposited by VA in Personal Funds of Patients that were derived from VA benefits where the veteran was inc...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] 38 CFR 12.4 (a) Upon authorization by the facility head or his or her designated representative, all funds, as defined in § 12.0 (except funds deposited by VA in Personal Funds of Patients that were derived from VA benefits where the veteran was incompetent at time of death)...	[ "kl3m:ecfr:80ff16ef81134fab", "kl3m:ecfrx:ecbdda73e373e3eb", "kl3m:ecfrx:8c11b4cdd0f120be", "kl3m:ecfrx:05d5c6e5f3b47863" ]	deterministic (opengloss-instruct v1)	564	false
og_gqa-3dc52e87f9316e13	og_gqa	grounded_qa	opengloss	[ { "role": "system", "content": "Be precise. If the sources do not contain the answer, say so." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] lycanthropy\nLycanthropy\n\nLycanthropy is a term used in mythology, folklore, psychiatry, and the history of medicine to denote the conditi...	<\|start\|><\|system\|> Be precise. If the sources do not contain the answer, say so.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] lycanthropy Lycanthropy Lycanthropy is a term used in mythology, folklore, psychiatry, and the history of medicine to denote the condition of becoming, or believing oneself to become, a wolf. In...	[ "og-qa:lycanthropy:noun:0#qa3", "og-enc:lycanthropy\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	355	false
kl3m_terms-3375c68bed8983b6	kl3m_terms	extraction	kl3m	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] § 12.1\n#### § 12.1 Definitions.\n\nAs used in this part, unless the context otherwise requires:\n\n(a) The term act means the act of August 1, 1956, Public Law 893, 84th Congress, 2d Session, requiring the registration of certain persons who have ...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] § 12.1 #### § 12.1 Definitions. As used in this part, unless the context otherwise requires: (a) The term act means the act of August 1, 1956, Public Law 893, 84th Congress, 2d Session, requiring the registration of certain persons who have knowledge of, or have received...	[ "kl3m:ecfrx:6b6e2a3c4667958b" ]	deterministic (opengloss-instruct v1)	245	true
wd_extract-442d40a7bed144b3	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Washington Irving\nWashington Irving: American writer, historian and diplomat (1783-1859). Washington Irving died in Tarrytown. Washington Irving died on November 28, 1859. Washington Irving was born on April 3, 1783.<\|/snippet\|>\n<\|/context\|>\nGive ...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Washington Irving Washington Irving: American writer, historian and diplomat (1783-1859). Washington Irving died in Tarrytown. Washington Irving died on November 28, 1859. Washington Irving was born on April 3, 1783.<\|/snippet\|> <\|/context\|> Give the fields name, born, leagu...	[ "wd:Q181667" ]	deterministic (opengloss-instruct v1)	174	true
og_gqa-8753a2a2267dc583	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] soil-based\nThe term is descriptive rather than strictly technical, but it has become conventional in fields such as agronomy, ecology, environmental science, and microbiology.\n\nIn agriculture and horticulture, soil-based cultivation refers to plan...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] soil-based The term is descriptive rather than strictly technical, but it has become conventional in fields such as agronomy, ecology, environmental science, and microbiology. In agriculture and horticulture, soil-based cultivation refers to plant production conducted in na...	[ "og-qa:soil_based:adjective:2#qa3", "og-qa:soil_based:adjective:2#qa6", "og-qa:soil_based:adjective:0#qa6", "og-enc:soil_based\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	452	false
og_rag-f6f9dd3ca9fda339	og_rag	citation	opengloss	[ { "role": "system", "content": "Use the numbered sources and say which one you used." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] civil servant\nA civil servant is a government employee who performs public duties without holding elected office. Civil servants work in ministries,...	<\|start\|><\|system\|> Use the numbered sources and say which one you used.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] civil servant A civil servant is a government employee who performs public duties without holding elected office. Civil servants work in ministries, agencies, and other public institutions, providing esse...	[ "og-qa:government_mandated:adjective:0#qa5", "og-enc:civil_servant\|college\|plain", "og-enc:boris_johnson\|neutral\|plain", "og-enc:government_mandated\|neutral\|plain", "og-enc:jail\|grade_1\|plain" ]	deterministic (opengloss-instruct v1)	630	false
og_abstain-04ed33e4151e34bd	og_abstain	abstention	opengloss	[ { "role": "user", "content": "Answer based on the source: What follows from the fact that anti-war protest during the Vietnam War era combined student activism, civil rights organizing, and artistic expression? Please be specific.\n<\|context\|>\n<\|snippet\|>[1] dominant authority\nDominant authority is the mo...	<\|start\|><\|user\|> Answer based on the source: What follows from the fact that anti-war protest during the Vietnam War era combined student activism, civil rights organizing, and artistic expression? Please be specific. <\|context\|> <\|snippet\|>[1] dominant authority Dominant authority is the most influential source of ru...	[ "og-qa:anti_war:noun:0#qa6", "og-enc:dominant_authority\|college\|plain" ]	deterministic (opengloss-instruct v1)	207	true
wd_extract-8db3d3f5f624cc26	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Going for the One\nGoing for the One: Album by Yes. Going for the One was released on the Atlantic Records label. Going for the One was published on July 22, 1977. Going for the One was performed by Yes.<\|/snippet\|>\n<\|/context\|>\nReturn YAML with th...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Going for the One Going for the One: Album by Yes. Going for the One was released on the Atlantic Records label. Going for the One was published on July 22, 1977. Going for the One was performed by Yes.<\|/snippet\|> <\|/context\|> Return YAML with the keys Name, Performer, Reco...	[ "wd:Q840277" ]	deterministic (opengloss-instruct v1)	163	true
wd_extract-4fe355a299a541ae	wd_extract	extraction	wikidata	[ { "role": "system", "content": "You are a helpful research assistant." }, { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Ubre Blanca\nUbre Blanca: Cuban dairy cow. Ubre Blanca was born in 1972. Ubre Blanca worked as a dairy cattle. Ubre Blanca died in 1985. Ubre Blanca lies in Cuba.<\|...	<\|start\|><\|system\|> You are a helpful research assistant.<\|eot\|> <\|user\|> <\|context\|> <\|snippet\|>[1] Ubre Blanca Ubre Blanca: Cuban dairy cow. Ubre Blanca was born in 1972. Ubre Blanca worked as a dairy cattle. Ubre Blanca died in 1985. Ubre Blanca lies in Cuba.<\|/snippet\|> <\|/context\|> Give me a JSON object with name,...	[ "wd:Q7876591" ]	deterministic (opengloss-instruct v1)	194	false
wd_extract-df3fe78a9bc62871	wd_extract	extraction	wikidata	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Gökçeören\nGökçeören is a village in Orta, Çankırı, Turkey. Gökçeören is on the continent of Asia. Gökçeören is in Turkey. Gökçeören is located in Orta.<\|/snippet\|>\n<\|/context\|>\nOutput a YAML mapping with these keys: name, officialLanguage, country...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Gökçeören Gökçeören is a village in Orta, Çankırı, Turkey. Gökçeören is on the continent of Asia. Gökçeören is in Turkey. Gökçeören is located in Orta.<\|/snippet\|> <\|/context\|> Output a YAML mapping with these keys: name, officialLanguage, country, locatedIn. Absent informat...	[ "wd:Q2692919" ]	deterministic (opengloss-instruct v1)	159	false
og_gqa-a34ea93518379174	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] Uppsala General Catalogue\nUGC numbers therefore function as catalogue addresses rather than as physical classifications.\n\nThe catalogue was important because it extended organized galaxy identification beyond the historically famous NGC and IC obj...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] Uppsala General Catalogue UGC numbers therefore function as catalogue addresses rather than as physical classifications. The catalogue was important because it extended organized galaxy identification beyond the historically famous NGC and IC objects. Its observations were ...	[ "og-qa:uppsala_general_catalogue:noun:0#qa9", "og-enc:uppsala_general_catalogue\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	249	false
og_gqa-63526250e0fa45a8	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] solenopsis\nSocial organization and reproduction: most species display eusocial colony structure with queens and workers; some taxa are polygynous (multiple queens) and exhibit varied mating systems, influencing colony dynamics and dispersal.\nEcolog...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] solenopsis Social organization and reproduction: most species display eusocial colony structure with queens and workers; some taxa are polygynous (multiple queens) and exhibit varied mating systems, influencing colony dynamics and dispersal. Ecology and behavior: ants in thi...	[ "og-qa:solenopsis:noun:0#qa6", "og-qa:solenopsis:noun:0#qa5", "og-enc:solenopsis\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	317	true
og_rag-2b30fdf3eac21616	og_rag	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] nondisjunction\nNondisjunction is the failure of chromosomes to segregate correctly during cell division. The term is used chiefly for meiosis, the specialized division that produces eggs and sperm. In normal meiosis, homologous chromosome pairs sepa...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] nondisjunction Nondisjunction is the failure of chromosomes to segregate correctly during cell division. The term is used chiefly for meiosis, the specialized division that produces eggs and sperm. In normal meiosis, homologous chromosome pairs separate during meiosis I, fol...	[ "og-qa:homo:noun:1#qa2", "og-enc:nondisjunction\|college\|plain", "og-enc:occupied_space\|grade_5\|plain", "og-enc:homo\|neutral\|plain", "og-enc:chromium\|neutral\|plain", "og-enc:analytical_chemistry\|grade_10\|plain" ]	deterministic (opengloss-instruct v1)	701	false
og_rag2-deab5ac61c3a9259	og_rag2	citation	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] urban greening\nUrban greening is the planned incorporation of vegetation and natural features into cities to improve environmental conditions, human well-being, and visual quality. It ranges from trees planted along streets to pocket parks, green ro...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] urban greening Urban greening is the planned incorporation of vegetation and natural features into cities to improve environmental conditions, human well-being, and visual quality. It ranges from trees planted along streets to pocket parks, green roofs, and living walls that...	[ "og-qa:synezesis:noun:0#qa0", "og-qa:record_of_events:noun:2#qa5", "og-enc:urban_greening\|college\|plain", "og-enc:g_i_taylor\|grade_5\|plain", "og-enc:synezesis\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	474	false
og_abstain-1b133ca1570d6820	og_abstain	abstention	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] cholesterosis cutis\nCholesterosis cutis is a skin problem. Fat builds up in the skin and makes many yellow lumps. Body cells take in too much fat and swell. The lumps can be bumps, flat spots, or bigger lumps. They often grow on elbows and knees. Th...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] cholesterosis cutis Cholesterosis cutis is a skin problem. Fat builds up in the skin and makes many yellow lumps. Body cells take in too much fat and swell. The lumps can be bumps, flat spots, or bigger lumps. They often grow on elbows and knees. They can also grow on hands,...	[ "og-qa:vena_angularis:noun:0#qa0", "og-enc:cholesterosis_cutis\|grade_1\|plain" ]	deterministic (opengloss-instruct v1)	348	true
og_gqa-2861e209c486a870	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "If still images in a sequence were not shown rapidly enough, what would be missing from the motion picture effect?\n<\|context\|>\n<\|snippet\|>[1] motion picture\nmotion picture\n\nThe term motion picture refers to a sequence of still images shown rapidly enough to create the illu...	<\|start\|><\|user\|> If still images in a sequence were not shown rapidly enough, what would be missing from the motion picture effect? <\|context\|> <\|snippet\|>[1] motion picture motion picture The term motion picture refers to a sequence of still images shown rapidly enough to create the illusion of movement. Also called...	[ "og-qa:motion_picture:noun:1#qa6", "og-qa:motion_picture:adjective:0#qa5", "og-enc:motion_picture\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	341	false
og_gqa-00dfef61159e66c0	og_gqa	grounded_qa	opengloss	[ { "role": "user", "content": "<\|context\|>\n<\|snippet\|>[1] political settlement\nThe late 20th and early 21st centuries saw more structured peace processes with international mediation and emphasis on inclusivity. Notable examples include the transition in South Africa (1994), the Good Friday Agreement in No...	<\|start\|><\|user\|> <\|context\|> <\|snippet\|>[1] political settlement The late 20th and early 21st centuries saw more structured peace processes with international mediation and emphasis on inclusivity. Notable examples include the transition in South Africa (1994), the Good Friday Agreement in Northern Ireland (1998), Nep...	[ "og-qa:political_settlement:noun:0#qa5", "og-enc:political_settlement\|neutral\|plain" ]	deterministic (opengloss-instruct v1)	274	true

End of preview. Expand in Data Studio

OpenGloss-Instruct

OpenGloss-Instruct is a grounded, verifiable instruction dataset in one chat format, built for the ALEA small language models. Every answer is either derived by code from a source passage that ships inside the example and checked by an independent verifier, or (the separate gpt6luna config) written by a language model and then checked programmatically against the passage.

It snowballs out from OpenGloss (encyclopedia passages and their 2.3M grounded QA pairs), Wikidata, the KL3M Data Project legal corpus (U.S. Code, eCFR, Federal Register, public laws, court opinions, EDGAR agreements; see the licence table for each), OpenLemma, and, for the gpt6luna config only, OpenGloss-Edu.

Related datasets:

OpenGloss: mjbommar/opengloss-v1.3-dictionary, -definitions, -query-examples-flat, -contrastive-examples, -hard-negative-pairs, -encyclopedia-variants;
the generator: mjbommar/opengloss-generator;
OpenGloss v2.4 (the release used here): mjbommar/opengloss-v2.4-lexicon, mjbommar/opengloss-v2.4-qa-pairs, mjbommar/opengloss-v2.4-encyclopedia, mjbommar/opengloss-v2.4-definitions;
OpenLemma: mjbommar/openlemma-v0.1-items.

Format

<|start|>[<|system|>\n...<|eot|>\n]<|user|>\n<|context|>\n<|snippet|>[1] title\npassage<|/snippet|>\n...<|/context|>\n
question<|eot|>\n<|assistant|>\nanswer[ <|cite|>n<|/cite|>][\n<|quote|>verbatim sentence<|/quote|>]<|eot|>\n...<|end|>

messages holds the conversation as {role, content} turns. Context, citation and quote markers appear inline in the content as the strings <|context|>, <|snippet|>, <|cite|> and <|quote|>.
text is the full rendering. In the ALEA tokenizer (alea-p1sw-staged-16k) every control token is a single special id. Remap them to your own chat template if you use another tokenizer.
There are no reasoning traces, no <|think|>.
Phrasings vary widely inside the one format: hundreds of instruction phrasings and system prompts. Every instruction opening, first instruction line, answer opening and system prompt is capped at ≤ 1% of documents (measured maxima: 0.54%, 0.22%, 0.59% and 0.48%).

Configs

config	documents	ALEA tokens	what
`deterministic` (default)	2,619,589	690,406,640	25 code-generated families below; $0
`gpt6luna`	96,785	33,270,866	natural questions with answers, verbatim quotes and abstentions; one-sentence / bullet / structured summaries; over short KL3M, OpenGloss, OpenGloss-Edu and Wikidata passages; written by gpt-6-luna, checked by code
total	2,716,374	723,677,506

Load one config with datasets.load_dataset("mjbommar/opengloss-instruct", "deterministic") or "gpt6luna". Both configs have the same columns: id, family, task, source, messages, text, seeds (seed ids), generator, n_tokens_alea (length in the ALEA tokenizer) and in_r2_decay.

Column in_r2_decay marks the 594,627 deterministic documents used in the ALEA R2 training run's decay (4% of its tokens).

Deterministic families

family	what	documents	ALEA tokens
`og_gqa`	One OpenGloss encyclopedia passage + one of OpenGloss's own grounded QA pairs; 35% add the verbatim supporting sentence, 30% add follow-up turns	434,739	140,941,619
`og_rag`	2-4 same-domain distractor passages + the answering passage; answer with a numbered citation (and quote)	184,789	121,894,002
`wd_extract_kshot`	Multi-turn: 1-3 extraction demos whose key sets and values never match the final query	166,912	88,425,780
`kl3m_which_section`	3-5 USC/eCFR sections titled by citation only; which one covers a heading (cite)	115,804	48,471,270
`wd_extract2`	As wd_extract with ~30% absent requested fields and a same-class distractor entity in the passage	168,764	40,210,031
`wd_extract`	Wikidata record verbalized; requested keys to JSON/YAML, absent keys null	220,042	38,644,103
`og_rag2`	Two questions over several passages, each answer cited; 25% have an unanswerable second question	68,652	36,689,147
`fmt_convert`	JSON / YAML / CSV conversion of record lists, round-trip checked	144,846	32,122,023
`og_closed`	Closed-book question naming its headword (OpenGloss QA pairs grounded in definitions/examples)	588,653	30,369,987
`og_rag_abstain`	2-4 passages, none answering: abstain	56,064	26,090,809
`og_abstain`	Question about one entry, passage about another that never names it: abstain	71,027	18,739,092
`wd_qa`	One fact question over a verbalized Wikidata record	153,317	16,845,796
`kl3m_rag_define`	Which source defines a term; verbatim quote of the defining clause + citation	24,892	13,165,393
`kl3m_define`	Quote the defining clause of a term from one provision	26,150	8,262,757
`kl3m_abstain_define`	The passages never contain the asked term: abstain	25,161	7,524,179
`kl3m_case_cites`	Every federal reporter citation (U.S., S. Ct., F., F.2d-4th, F. Supp.) in an opinion excerpt, JSON array	19,501	7,386,969
`kl3m_terms`	Every defined term in a definitions excerpt, JSON array	13,174	4,008,189
`og_mcq`	Which word matches a definition (4 options)	51,455	3,106,570
`og_define`	Define a word (OpenGloss glossary)	57,138	2,473,935
`kl3m_cites`	Every U.S.C./CFR citation in order, as a JSON array	6,338	1,774,855
`fr_extract`	Federal Register header fields to JSON/YAML (contact sections removed)	2,882	1,347,335
`fr_qa`	One header-field question over a Federal Register document	2,881	1,136,103
`ol_dates`	OpenLemma date arithmetic items, chat rendering	6,590	284,345
`ol_syllogism`	OpenLemma syllogism validity items, chat rendering	4,059	266,873
`ol_arith`	OpenLemma arithmetic items, chat rendering	5,759	225,478

By task: grounded_qa 617,087, extraction 597,613, closed_qa 588,653, citation 394,137, abstention 152,252, conversion 144,846, definition 108,593, procedural 16,408 documents. By source (tokens): OpenGloss 55.1%, Wikidata 31.3%, KL3M 13.5%, OpenLemma 0.1%.

`gpt6luna` families

family	source	documents	ALEA tokens
`llm_qa`	KL3M legal	18,443	6,600,110
`llm_qa`	OpenGloss encyclopedia	14,028	5,370,418
`llm_qa`	OpenGloss-Edu	10,146	2,914,492
`llm_qa`	Wikidata	8,089	1,869,091
`llm_summary`	KL3M legal	16,395	6,150,720
`llm_summary`	OpenGloss encyclopedia	12,843	5,158,034
`llm_summary`	OpenGloss-Edu	9,261	2,909,041
`llm_summary`	Wikidata	7,580	2,298,960

By source (tokens): KL3M legal 38.3%, OpenGloss encyclopedia 31.6%, OpenGloss-Edu 17.5%, Wikidata 12.5%. The source mix follows the generation plan (passages sampled roughly 35% KL3M, 25% OpenGloss, 25% OpenGloss-Edu, 15% Wikidata); it was not rebalanced after the run.

How it was checked

Independent verifier. A separate module re-derives every answer from the seed data, looked up by id, and from the rendered text. It does not import the generators. It checks:
- every passage is a verbatim substring of its seed;
- grounded answers have content-word support in their passage, and quotes are verbatim;
- a citation must point at the only passage from the right entry, which must be the best-supported one;
- abstention passages never name the subject or support its answer;
- Wikidata values are re-read from the record, and a field is null exactly when the passage does not state it for the target entity;
- Federal Register fields are re-parsed;
- definition quotes must be the term's defining clause;
- citation lists must cover every citation anchor in the passage, in order;
- conversions must round-trip;
- date and arithmetic answers are recomputed, and syllogism validity is decided by model enumeration.
Result. Every released document passes. A negative control corrupts 1 in 5 documents (wrong citation, swapped answer, changed or nulled field, dropped or extra list item, answering passage swapped into an abstention item, altered quote, injected number or entity). 100% of the corrupted copies are rejected (517,758 of 517,758).
The gpt6luna config was checked the same way in separate processes: 104,210 of 104,233 rendered documents pass (the 23 that do not were dropped), and 20,932 of 20,932 corrupted copies are rejected (injected number or name, a repeated answer, print-header noise put back into the passage).
Decontamination (fail closed) removed 12,145 deterministic and 2,579 gpt6luna documents. It checked against the ALEA evaluation suites: likelihood benchmarks (ARC, SciQ, LAMBADA, HellaSwag, PIQA, WinoGrande, BoolQ, OBQA, MMLU) by 13-gram; SHELF validation/test; knowledge and general-facts probes; ICL suites; out-of-format ICL; judge and template-leak prompts; the OpenLemma eval release; a structured-extraction baseline; and a held-out chat-format probe. Probe seeds are also held out by id.
Personal data. No private contact details are added by the generators. Names in the public legal record (parties, judges, officials) are kept.
- fr_extract / fr_qa passages are cut before the Federal Register's FOR FURTHER INFORMATION CONTACT section.
- gpt6luna: passages with an e-mail address or a phone number were dropped before release, and the verifier rejects any such pattern in a passage or an answer.
- deterministic: some legal passages still contain official contact points printed in the source documents (agency mailboxes such as FOIA addresses, and office phone numbers in the Federal Register and eCFR).

The `gpt6luna` config

Generator: OpenAI gpt-6-luna (flex tier, low reasoning effort), with one short passage (≤ ~1,200 characters) per call. Asker profile, style and question types are sampled in code, and 20% of calls request one unanswerable question. 53,085 calls were made (a 2,745-call pilot and a main run stopped at about 50K calls), for a total generation cost of $11.09.
Programmatic checks on every item before rendering:
- quotes must be verbatim;
- answers need ≥ 60% stem support;
- every number in an answer or summary must appear in the passage, and so must every capitalized name (a sentence-initial word may instead match a source word by prefix);
- summary sentences need ≥ 70% support;
- structured-summary entities must be verbatim.
Rendering: a QA document per call (every passing question as one turn; 30% of answers add the verbatim quote) and a summary document per call (every passing summary form as one turn). Unanswerable questions get an abstention that says only that the source lacks the answer.
Cleanups at render time:
- Federal Register / public-law print headers (page headers, GPO user stamps, VerDate/Jkt/Frm/Fmt/Sfmt markers) were stripped from 295 passages; questions, quotes and summaries that depended on the removed text were dropped (52 questions, 31 summaries);
- 1,364 near-duplicate answers within a document (token Jaccard ≥ 0.8) were dropped;
- 57 passages with contact details and 5 with control characters were dropped.
Release caps: documents are admitted only while every instruction opening, first instruction line, answer opening and system prompt stays under 1% of the config (4,846 documents skipped).
Every record keeps its seed id. The generator is disclosed here; check OpenAI's terms for your own use of model outputs.

Sources and licences

source	licence / status	used as
OpenGloss v2.4 (lexicon encyclopedia texts, qa-pairs, glossary)	the author's own dataset, CC-BY-4.0; WordNet-derived rows under the Princeton WordNet License (notice below and `LICENSE-WordNet.txt`)	passages, questions, answers, glossary
OpenGloss-Edu v1 (not yet published)	the author's own synthetic corpus, generated with `gpt-6-luna`; released here under CC-BY-4.0	passages for the `gpt6luna` config
OpenLemma v0.1 items (`mjbommar/openlemma-v0.1-items` @ d0b6aa57)	the author's own dataset, CC-BY-4.0 (WordNet notice applies to WordNet-derived rows)	dates / arithmetic / syllogism items (train split, verified tier)
Wikidata (2025-11 dump)	CC0 1.0	records and facts
U.S. Code, public laws, eCFR, Federal Register (via the KL3M Data Project)	U.S. government works, not subject to copyright (17 U.S.C. § 105)	passages
Court opinions (Caselaw Access Project, via the KL3M Data Project)	judicial opinions are not copyrightable (government edicts doctrine); CAP data released by Harvard Law School Library under CC0 1.0	passages
EDGAR agreements (via the KL3M Data Project, `alea-institute/kl3m-data-edgar-agreements`)	public SEC filings (exhibits) made available on EDGAR for unrestricted reuse; distributed by the KL3M Data Project under CC BY 4.0	short excerpts (definitions sections and other windows)
Generated text (instructions, renderings, `gpt6luna` answers and summaries)	CC-BY-4.0	everything else

The dataset as a whole is released under CC-BY-4.0. Please attribute OpenGloss-Instruct and OpenGloss.

WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved. (Full licence text, including the disclaimer, in LICENSE-WordNet.txt.)

SHELF is not used anywhere in this dataset, as seed or content. Its validation and test splits are only a decontamination target.

Intended use

Supervised or mid-training data for small language models: grounded answering with citations and verbatim quotes, abstention when the source lacks the answer, extraction to JSON/YAML, format conversion, and short summaries.
2,048-token friendly: every document is ≤ 2,048 tokens in the ALEA tokenizer (n_tokens_alea), so documents fit a 2K context without truncation.
Sample by family / task / source to build a mix. deterministic and gpt6luna can be used separately, for example to measure the value of the LLM-written part.

Limitations

The deterministic families are template-rendered. Phrasings are diverse, but the answers are as stylistically narrow as their sources: OpenGloss's answers, Wikidata's verbalizations, and statutory text.
Wikidata verbalizations are simple sentences and can carry dump errors, such as a mislabeled description.
"Unanswerable" means not answerable from the provided passages. Some such questions are answerable from general knowledge.
Legal passages include OCR artifacts from the source corpora.
Shares are uneven: the extraction families are token-heavy. Sample by family / task for a balanced mix.
No reasoning traces. Answers are direct; there is no chain-of-thought or <|think|> content.
Deterministic vs LLM parts. The deterministic answers are derived by code and verified exactly. The gpt6luna answers and summaries are written by a model and checked by heuristics (verbatim quotes, word support, numbers and names present in the passage). The checks catch added facts but not every paraphrase error or omission.
The gpt6luna config is small (one passage per call, about 53K calls) and its source mix was not rebalanced.

Citation

@misc{bommarito2025opengloss,
  title  = {OpenGloss: A Synthetic Encyclopedic Dictionary and Semantic Knowledge Graph},
  author = {Bommarito, Michael J., II},
  year   = {2025},
  eprint = {2511.18622},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2511.18622}
}