Dataset Viewer
Auto-converted to Parquet Duplicate
id
string
text
string
family
string
kind
string
source
string
source_id
string
n_tasks
int64
schema_styles
string
keys
string
instructions
string
gold
string
n_tokens
int64
ej3-p00000-05e234e1d6-t2
<|start|><<< Death This Agreement shall terminate immediately upon the death of the Executive. In such event, Employer shall pay Executive’s estate an amount equal to 12 months Base Salary, such amount being payable within three months after his death. 2 --- Disability This Agreement shall not terminate upon the tem...
extract_json_v3
plain
edgar_contract
0001079973-08-001097
1
["skeleton_example"]
[["cause_items", "hq_city", "notice_term", "confidentiality_section"]]
["Extract the Cause and notice details from this agreement excerpt as JSON matching the skeleton: {\"cause_items\":[{\"label\":\"\",\"text\":\"\"}],\"hq_city\":\"\",\"notice_term\":\"\",\"confidentiality_section\":\"\"}. cause_items lists each lettered Cause event in order: label is the letter in parentheses as printed...
[{"cause_items": [{"label": "a", "text": "A finding by a court of breach of fiduciary duty or conviction of criminal conduct by the Executive, in either case after all rights of appeal have expired or such appeals have been exhausted, having the effect of materially adversely affecting the Company and/or its reputation...
1,213
ej3-p00000-25eb365ce3-t1
<|start|><|user|> From this employment agreement excerpt, pull out the dispute-resolution details and the prior employment issue. Return JSON with these keys: - governing_state: the name of the state whose laws govern the agreement, just the state name as written (no 'State of'). - venue: an object with state_court_cou...
extract_json_v3
chat
edgar_contract
0000950152-94-001159
1
["key_list_with_descriptions"]
[["governing_state", "venue", "former_employer"]]
["From this employment agreement excerpt, pull out the dispute-resolution details and the prior employment issue. Return JSON with these keys:\n- governing_state: the name of the state whose laws govern the agreement, just the state name as written (no 'State of').\n- venue: an object with state_court_county (the count...
[{"governing_state": "Ohio", "venue": {"state_court_county": "Cuyahoga County, Ohio", "federal_district": "Northern District of Ohio, Eastern Division", "federal_courthouse_city": "Cleveland, Ohio"}, "former_employer": "Basicomputer Corporation"}]
753
ej3-p00000-25eb365ce3-t2
<|start|><|system|> You are a careful assistant. Use the text you are given.<|eot|> <|user|> SEVERABLE PROVISIONS The provisions of this Agremeent are severable, and if any one or more provisions may be determined to be illegal or otherwise unenforceable, in whole or in part, the remaining provisions and any partially ...
extract_json_v3
chat_sys
edgar_contract
0000950152-94-001159
1
["skeleton_example"]
[["signatories", "section_headings", "indemnified_company"]]
["I need the execution block and the clause structure of this agreement excerpt. Fill in this skeleton (keep the key names):\n{\"signatories\": [{\"name\": \"<printed name of each person who signs, exactly as typed under the signature line>\", \"title\": \"<that person's office title as printed, or null if none is prin...
[{"signatories": [{"name": "Dan R. Wipff", "title": "President and Chief Operating Officer", "signing_for": "TELXON CORPORATION"}, {"name": "FRANK BRICK", "title": null, "signing_for": null}], "section_headings": ["SEVERABLE PROVISIONS", "GOVERNING LAW AND VENUE", "FORMER EMPLOYER", "SURVIVAL"], "indemnified_company": ...
824
ej3-p00001-1e95e7979a-t2
<|start|>Compensation Upon Termination (a) If the Executive’s employment is terminated by the Company for Disability or Cause or by the Executive other than for Good Reason, the Company shall have no obligation to pay any compensation to the Executive under this Agreement in respect of periods beginning on or after the...
extract_json_v3
plain
edgar_contract
0000950137-03-005852
1
["prose"]
[["floor_amounts", "bonus_installment_phrase", "early_commencement_age", "dispute_section"]]
["Read the severance provisions and give me a JSON object with four keys. floor_amounts is a list of the dollar figures that act as minimums or fixed amounts in the section, each written with its dollar sign and number as printed and listed once in order of first appearance. bonus_installment_phrase is the short phrase...
[{"floor_amounts": ["$175,000", "$12,000"], "bonus_installment_phrase": "two equal installments", "early_commencement_age": "age 65", "dispute_section": "Section 1(e)"}]
1,826
ej3-p00001-285cc8fb79-t12
<|start|>Here is one. This is a Form 8-K. List the items reported and identify the registrant. Return an object with: registrantName (name as printed in capitals), commissionFileNumber, incorporationState, eventDate (the date of earliest event reported), and items (each reported item as an object with itemNumber, e.g. ...
extract_json_v3
plain
edgar_8k
0000018255-10-000009
2
["prose", "json_schema"]
[["registrantName", "commissionFileNumber", "incorporationState", "eventDate", "items"], ["signers", "signatureDate", "fiscalPeriodEnd", "exhibitNumber", "principalOfficeAddress"]]
["This is a Form 8-K. List the items reported and identify the registrant. Return an object with: registrantName (name as printed in capitals), commissionFileNumber, incorporationState, eventDate (the date of earliest event reported), and items (each reported item as an object with itemNumber, e.g. the number only like...
[{"registrantName": "THE CATO CORPORATION", "commissionFileNumber": "1-31340", "incorporationState": "Delaware", "eventDate": "March 18, 2010", "items": [{"itemNumber": "2.02", "itemTitle": "Results of Operations and Financial Condition"}, {"itemNumber": "9.01", "itemTitle": "Financial Statements and Exhibits"}]}, {"si...
1,068
ej3-p00002-49e20bffd2-t12
<|start|><|user|> Pull the basic case information from this docket sheet. Keys: caption (case name as printed), docket_number (as printed), assigned_judge (name as printed), date_filed (as printed), date_terminated (as printed), cause (the statute or cause as printed after 'Cause:'). LONG v. ENDOCRINE SOCIETY Docket N...
extract_json_v3
chat
court_docket
1:16-cv-00026
2
["key_list_with_descriptions", "prose"]
[["caption", "docket_number", "assigned_judge", "date_filed", "date_terminated", "cause"], ["scheduling_deadlines", "summaryJudgmentBriefing", "mediationClerk", "proHacViceAttorney"]]
["Pull the basic case information from this docket sheet. Keys: caption (case name as printed), docket_number (as printed), assigned_judge (name as printed), date_filed (as printed), date_terminated (as printed), cause (the statute or cause as printed after 'Cause:').", "I'm tracking the schedule in this case. Give me ...
[{"caption": "LONG v. ENDOCRINE SOCIETY", "docket_number": "1:16-cv-00026", "assigned_judge": "Amit P. Mehta", "date_filed": "2016-01-07", "date_terminated": "2017-07-14", "cause": "Family and Medical Leave Act of 1993"}, {"scheduling_deadlines": [{"milestone": "joinder of parties and amendment of pleadings", "due": "J...
2,242
ej3-p00002-ce8a94e3d1-t12
<|start|>Summarize the header of this court docket as JSON matching this schema: {"type":"object","properties":{"case_info":{"type":"object","description":"header facts","properties":{"case_name":{"type":"string","description":"the caption exactly as written on the first line"},"docket_number":{"type":"string","descrip...
extract_json_v3
plain
court_docket
4:17-cv-01393
2
["json_schema", "key_list_with_descriptions"]
[["case_info", "lifecycle"], ["counsel_appearances", "magistrate_judge", "service"]]
["Summarize the header of this court docket as JSON matching this schema: {\"type\":\"object\",\"properties\":{\"case_info\":{\"type\":\"object\",\"description\":\"header facts\",\"properties\":{\"case_name\":{\"type\":\"string\",\"description\":\"the caption exactly as written on the first line\"},\"docket_number\":{\...
[{"case_info": {"case_name": "Figueroa v. Everalbum, Inc.", "docket_number": "4:17-cv-01393", "assigned_judge": "Ronnie L. White", "cause": "47:0227 Telephone Consumer Protection Act", "nature_of_suit": "890 Other Statutory Actions"}, "lifecycle": {"date_filed": "2017-04-26", "date_terminated": "2017-10-12"}}, {"counse...
2,179
ej3-p00003-2e34063219-t1
<|start|><|system|> You are a concise language assistant.<|eot|> <|user|> This excerpt covers how a stock plan can be amended and when it takes effect. Please return JSON with: amending_body (the name of the body that may amend, suspend or terminate the plan, without 'The'); option_type_with_price_floor (the type of op...
extract_json_v3
chat_sys
edgar_contract
0000950134-95-000310
1
["prose"]
[["amending_body", "option_type_with_price_floor", "adoption_vote"]]
["This excerpt covers how a stock plan can be amended and when it takes effect. Please return JSON with: amending_body (the name of the body that may amend, suspend or terminate the plan, without 'The'); option_type_with_price_floor (the type of option, as written, whose minimum Option Price may not be decreased withou...
[{"amending_body": "Board", "option_type_with_price_floor": "Incentive Option", "adoption_vote": {"approving_body": "shareholders", "threshold": "a majority of the votes cast", "meeting": "1994 Annual Meeting"}}]
467
ej3-p00003-2e34063219-t2
<|start|>Quick question: extract these items from the plan excerpt, using exactly these keys: {"section_headings": [<each section heading as printed>], "counsel_alternative": <the short phrase naming the document from counsel that can substitute for stockholder approval of Incentive Option changes>, "protected_particip...
extract_json_v3
plain
edgar_contract
0000950134-95-000310
1
["skeleton_example"]
[["section_headings", "counsel_alternative", "protected_participant_groups", "holder_protection_phrase"]]
["Extract these items from the plan excerpt, using exactly these keys: {\"section_headings\": [<each section heading as printed>], \"counsel_alternative\": <the short phrase naming the document from counsel that can substitute for stockholder approval of Incentive Option changes>, \"protected_participant_groups\": [<ea...
[{"section_headings": ["Amendment or Termination of Plan", "Adoption of the Plan"], "counsel_alternative": "opinion of counsel", "protected_participant_groups": ["Executive Officers", "Outside Directors"], "holder_protection_phrase": "consent of the holder"}]
449
ej3-p00003-389bae58fd-t1
<|start|>""" BURRELL v. JOHNSON & JOHNSON Docket No. 3:17-cv-13642 Assigned to: Freda L. Wolfson Referred to: Lois H. Goodman Date filed: 2017-12-27 Cause: 28:1332 Diversity-Product Liability Nature of suit: 367 Personal Injury: Health Care/Pharmaceutical Personal Injury Product Liability Docket entries: 1 2017-12-27...
extract_json_v3
plain
court_docket
3:17-cv-13642
1
["key_list_with_descriptions"]
[["caption", "docketNumber", "districtJudge", "magistrateJudge", "dateFiled", "causeOfAction"]]
["Summarize the docket header for this case as JSON: caption, docketNumber, districtJudge, magistrateJudge (names as printed), dateFiled (as printed), and causeOfAction (as printed after 'Cause:')."]
[{"caption": "BURRELL v. JOHNSON & JOHNSON", "docketNumber": "3:17-cv-13642", "districtJudge": "Freda L. Wolfson", "magistrateJudge": "Lois H. Goodman", "dateFiled": "2017-12-27", "causeOfAction": "28:1332 Diversity-Product Liability"}]
1,295
ej3-p00003-389bae58fd-t2
<|start|>Who is suing whom, and who is representing them? Return: plaintiff (name as printed in the complaint entry), defendants (list of the defendant names from the complaint entry, each as printed, without the parenthetical nickname), mdlNumber (the MDL number as written), and appearances (list of objects, one per n...
extract_json_v3
plain
court_docket
3:17-cv-13642
1
["prose"]
[["plaintiff", "defendants", "mdlNumber", "appearances"]]
["Who is suing whom, and who is representing them? Return: plaintiff (name as printed in the complaint entry), defendants (list of the defendant names from the complaint entry, each as printed, without the parenthetical nickname), mdlNumber (the MDL number as written), and appearances (list of objects, one per notice o...
[{"plaintiff": "MANUELA BURRELL", "defendants": ["IMERYS TALC AMERICA, INC", "JOHNSON & JOHNSON", "JOHNSON & JOHNSON CONSUMER, INC.", "PERSONAL CARE PRODUCT COUNCIL"], "mdlNumber": "16-2738", "appearances": [{"attorney": "DOUGLAS ROBERT PLYMALE", "onBehalfOf": "MANUELA BURRELL"}, {"attorney": "SUSAN M. SHARKO", "onBeha...
1,675
ej3-p00004-7bf390c816-t2
<|start|>For the administration part of this plan, give me an object with: administrator (the defined body that administers the Plan, one word), boardOverrideSection (the section reference where the Board may exercise Committee powers), quorumRule (the full sentence stating what constitutes a quorum, as written), writt...
extract_json_v3
plain
edgar_contract
0000910647-98-000114
1
["prose"]
[["administrator", "boardOverrideSection", "quorumRule", "writtenConsentCondition", "sarSection"]]
["For the administration part of this plan, give me an object with: administrator (the defined body that administers the Plan, one word), boardOverrideSection (the section reference where the Board may exercise Committee powers), quorumRule (the full sentence stating what constitutes a quorum, as written), writtenConse...
[{"administrator": "Committee", "boardOverrideSection": "Section 2(e)", "quorumRule": "A majority\nof the members of the Committee shall constitute a quorum, and the action of a\nmajority of the members present at any meeting at which a quorum is present\nshall be deemed the action of the Committee.", "writtenConsentCo...
1,158
ej3-p00004-886241d603-t1
<|start|>This is the miscellaneous and tax-compliance part of an executive agreement. Return: - guidance_cited: a list of the IRS notice and the revenue procedure named in the text, each exactly as written, in document order. - payment_delay: an object describing the delay for certain employees, with employee_category ...
extract_json_v3
plain
edgar_contract
0001144204-17-026076
1
["key_list_with_descriptions"]
[["guidance_cited", "payment_delay"]]
["This is the miscellaneous and tax-compliance part of an executive agreement. Return:\n- guidance_cited: a list of the IRS notice and the revenue procedure named in the text, each exactly as written, in document order.\n- payment_delay: an object describing the delay for certain employees, with employee_category (the ...
[{"guidance_cited": ["IRS Notice 2010-6", "Rev. Proc. 92-64"], "payment_delay": {"employee_category": "specified employee", "delay_period": "six-month period", "trust_authority": "Rev. Proc. 92-64"}}]
848
ej3-p00004-886241d603-t2
<|start|><|system|> You are a careful analyst. Be accurate.<|eot|> <|user|> Give me JSON with these keys. section_headings: the three unnumbered clause headings at the top of the excerpt, as printed. reimbursement_rules: the three numbered conditions (i), (ii) and (iii) governing expense reimbursement under Section 409...
extract_json_v3
chat_sys
edgar_contract
0001144204-17-026076
1
["prose"]
[["section_headings", "reimbursement_rules", "counterpart_phrase", "withholding_sentence"]]
["Give me JSON with these keys. section_headings: the three unnumbered clause headings at the top of the excerpt, as printed. reimbursement_rules: the three numbered conditions (i), (ii) and (iii) governing expense reimbursement under Section 409A, each as the clause text after its numeral, without the numeral, parenth...
[{"section_headings": ["Severability", "Counterparts", "Blue Penciling"], "reimbursement_rules": ["in no event shall any reimbursement be paid after the last day of the taxable year following the taxable year in which the expense was incurred", "the amount of reimbursable expenses incurred or provision of in-kind benef...
908
ej3-p00005-b5ad1be498-t2
<|start|>``` WARRANTIES, REPRESENTATIONS AND COVENANTS WARRANTIES, REPRESENTATIONS AND COVENANTS BEASLEY Beasley represents and warrants that: (a) it has full power and authority to enter into and fully perform this Agreement; and (b) this Agreement has been duly authorized and is...
extract_json_v3
plain
edgar_contract
0000912057-00-015321
1
["key_list_with_descriptions"]
[["indemnityPairs", "indemnifiedPartyTerm", "indemnitorTerm", "coveredSections", "exclusiveRemedyArticle"]]
["Extract the indemnification mechanics as JSON. Keys: indemnityPairs (list of objects, one per indemnity, with indemnitor = the party that indemnifies as written, and indemnitees = the defined term for those protected, as written inside the quotation marks), indemnifiedPartyTerm and indemnitorTerm (the defined labels ...
[{"indemnityPairs": [{"indemnitor": "Findwhat", "indemnitees": "Beasley Indemnitees"}, {"indemnitor": "Beasley", "indemnitees": "FINDWHAT INDEMNITEES"}], "indemnifiedPartyTerm": "INDEMNIFIED PARTY", "indemnitorTerm": "INDEMNITOR", "coveredSections": "Sections 5.1 or 5.2", "exclusiveRemedyArticle": "Article 5"}]
1,198
ej3-p00005-d5ff92e71a-t1
<|start|>From this stock option plan excerpt, extract: - plan_title: the plan's name as written in the Purposes section (without 'this'). - company: an object with name (the company's full name as written in its definition), defined_term (the short defined word used for it, without quotes), and country (the single word...
extract_json_v3
plain
edgar_contract
0000891618-01-502225
1
["key_list_with_descriptions"]
[["plan_title", "company", "share_pool", "option_character"]]
["From this stock option plan excerpt, extract:\n- plan_title: the plan's name as written in the Purposes section (without 'this').\n- company: an object with name (the company's full name as written in its definition), defined_term (the short defined word used for it, without quotes), and country (the single word nami...
[{"plan_title": "2001 Outside Director Option Plan", "company": {"name": "SmartForce Public Limited Company", "defined_term": "Company", "country": "Ireland"}, "share_pool": "350,000 Shares", "option_character": "nonstatutory stock options"}]
1,032
ej3-p00006-251ad859c6-t1
<|start|><|user|> """ Governing Law All questions arising with respect to the provisions of this Award Agreement shall be determined by application of the laws of Oklahoma, without giving any effect to any conflict of law provisions thereof, except to the extent Oklahoma state law is preempted by federal law. The oblig...
extract_json_v3
chat
edgar_contract
0000732834-18-000009
1
["key_list_with_descriptions"]
[["governing_law_state", "venue_courts", "company"]]
["This is an excerpt from an equity award agreement. Pull out the following and return a JSON object with exactly these keys:\n- governing_law_state: the name of the state whose laws govern, as written.\n- venue_courts: list of the courts that are given jurisdiction and venue, each court description as written (e.g. 's...
[{"governing_law_state": "Oklahoma", "venue_courts": ["state courts located in Oklahoma County, Oklahoma", "United States District Court for the Western District of Oklahoma"], "company": {"name": "CONTINENTAL RESOURCES, INC.", "entity_description": "Oklahoma corporation"}}]
753
ej3-p00006-251ad859c6-t2
<|start|><|system|> You are a concise tutor. Give only what is asked.<|eot|> <|user|> From this award agreement excerpt, I need the tax acknowledgment details and the amendment mechanics. Return JSON with an `acknowledgment` object (released_group_term: the defined term naming the group the participant releases, withou...
extract_json_v3
chat_sys
edgar_contract
0000732834-18-000009
1
["prose"]
[["acknowledgment", "amendment_decision_makers", "consent_party"]]
["From this award agreement excerpt, I need the tax acknowledgment details and the amendment mechanics. Return JSON with an `acknowledgment` object (released_group_term: the defined term naming the group the participant releases, without quotes; tax_election: the Code section of the tax election the participant is told...
[{"acknowledgment": {"released_group_term": "Company Parties", "tax_election": "section 83(b)", "election_deadline": "30 days after the Date of Grant"}, "amendment_decision_makers": ["Board", "Committee"], "consent_party": "Participant"}]
792
ej3-p00006-ff3817c8fa-t1
<|start|><|user|> From this federal docket sheet, pull the case identification block. Return JSON with these keys: - caption: the case name exactly as printed at the top. - docketNumber: the docket number as written after 'Docket No.'. - judges: an object with district_judge (the name of the judge listed after 'Assigne...
extract_json_v3
chat
court_docket
2:17-cv-10368
1
["key_list_with_descriptions"]
[["caption", "docketNumber", "judges", "dateTerminated", "cause"]]
["From this federal docket sheet, pull the case identification block. Return JSON with these keys:\n- caption: the case name exactly as printed at the top.\n- docketNumber: the docket number as written after 'Docket No.'.\n- judges: an object with district_judge (the name of the judge listed after 'Assigned to:') and m...
[{"caption": "Ferland v. BERRYHILL", "docketNumber": "2:17-cv-10368", "judges": {"district_judge": "Mark A. Goldsmith", "magistrate_judge": "Stephanie Dawkins Davis"}, "dateTerminated": "2018-08-28", "cause": "42:205 Denial Social Security Benefits"}]
1,518
ej3-p00006-ff3817c8fa-t2
<|start|>""" Ferland v. BERRYHILL Docket No. 2:17-cv-10368 Assigned to: Mark A. Goldsmith Referred to: Stephanie Dawkins Davis Date filed: 2017-02-06 Date terminated: 2018-08-28 Cause: 42:205 Denial Social Security Benefits Nature of suit: 863 Social Security: DIWC/DIWW Docket entries: 1 2017-02-06 COMPLAINT filed b...
extract_json_v3
plain
court_docket
2:17-cv-10368
1
["prose"]
[["plaintiff", "appearances", "scheduleDeadlines", "extensionMotionTitle"]]
["I'm tracking the briefing in this Social Security appeal. Give me a JSON object with: plaintiff (the plaintiff's name as written in the complaint entry); appearances (a list of objects, one per notice of appearance, with attorney = the attorney's name as printed and client = the party name the attorney appeared for, ...
[{"plaintiff": "Daniel R. Ferland", "appearances": [{"attorney": "Timothy S. Bolen", "client": "NANCY A BERRYHILL"}], "scheduleDeadlines": {"plaintiffMotionDue": "8/9/2017", "defendantResponseDue": "9/8/2017", "plaintiffReplyDue": "9/22/2017"}, "extensionMotionTitle": "Ex Parte MOTION TO EXTEND Time to File Plaintiff's...
1,524
ej3-p00007-ae191cfed0-t12
<|start|><|system|> You are a concise assistant. Follow the requested format.<|eot|> <|user|> Summarize the header of this court docket as JSON shaped like {"caption": ..., "docket_number": ..., "judges": {"assigned_to": ..., "district_judge": ...}, "date_filed": ..., "date_terminated": ...}. caption = case name as pri...
extract_json_v3
chat_sys
court_docket
4:17-cv-00001
2
["skeleton_example", "json_schema"]
[["caption", "docket_number", "judges", "date_filed", "date_terminated"], ["plaintiff", "cause", "nature_of_suit", "motions"]]
["Summarize the header of this court docket as JSON shaped like {\"caption\": ..., \"docket_number\": ..., \"judges\": {\"assigned_to\": ..., \"district_judge\": ...}, \"date_filed\": ..., \"date_terminated\": ...}. caption = case name as printed on the first line; docket_number = the number after 'Docket No.'; assigne...
[{"caption": "Blumenberg v. Commissioner of Social Security", "docket_number": "4:17-cv-00001", "judges": {"assigned_to": "Jane M. Virden", "district_judge": "Sharion Aycock"}, "date_filed": "2017-01-03", "date_terminated": "2018-02-22"}, {"plaintiff": "Patricia Blumenberg", "cause": "42:405 Review of HHS Decision (SSI...
1,479
ej3-p00007-e97595cb5c-t1
<|start|>Extract the cover-page facts from this Form 8-K using this schema: {"type":"object","properties":{"registrantName":{"type":"string","description":"registrant's exact name as printed on the cover, including series"},"incorporationState":{"type":"string","description":"state of incorporation as printed"},"commis...
extract_json_v3
plain
edgar_8k
0000882377-03-000490
1
["json_schema"]
[["registrantName", "incorporationState", "commissionFileNumber", "reportDate", "itemHeading"]]
["Extract the cover-page facts from this Form 8-K using this schema: {\"type\":\"object\",\"properties\":{\"registrantName\":{\"type\":\"string\",\"description\":\"registrant's exact name as printed on the cover, including series\"},\"incorporationState\":{\"type\":\"string\",\"description\":\"state of incorporation as...
[{"registrantName": "ABN AMRO Mortgage Corporation, Series 2003-3", "incorporationState": "DELAWARE", "commissionFileNumber": "333-101550", "reportDate": "February 27, 2003", "itemHeading": "Item 2. Acquisition or Disposition of Assets."}]
1,334
ej3-p00008-8e28f42fde-t1
<|start|>[[[ SECURITIES AND EXCHANGE COMMISSIONWashington, D.C. 20549 FORM 8-K Current Report Pursuant to Section 13 or 15(d)of the Securities Exchange Act of 1934 Date of Report (Date of earliest event reported): January 17, 2008 |TRANS WORLD ENTERTAINMENT CORPORATION |------------...
extract_json_v3
plain
edgar_8k
0000930413-08-000398
1
["key_list_with_descriptions"]
[["registrantName", "stateOfIncorporation", "eventDate", "items"]]
["Pull the basics from this Form 8-K.\nregistrantName: the company name written in sentence case in the body text (e.g. as it appears when describing the press release), without 'the'.\nstateOfIncorporation: the state named as jurisdiction of incorporation.\neventDate: the date of the event reported, as written.\nitems...
[{"registrantName": "Trans World Entertainment Corporation", "stateOfIncorporation": "New York", "eventDate": "January 17, 2008", "items": [{"item_number": "8.01", "item_title": "OTHER EVENTS"}, {"item_number": "9.01", "item_title": "FINANCIAL STATEMENTS AND EXHIBITS"}]}]
939
ej3-p00008-8e28f42fde-t2
<|start|><|system|> You are an attentive editor. Answer briefly.<|eot|> <|user|> """ SECURITIES AND EXCHANGE COMMISSIONWashington, D.C. 20549 FORM 8-K Current Report Pursuant to Section 13 or 15(d)of the Securities Exchange Act of 1934 Date of Report (Date of earliest event reported): January 17, 2008 |TRANS WORLD ...
extract_json_v3
chat_sys
edgar_8k
0000930413-08-000398
1
["prose"]
[["closing_locations", "press_release_exhibit", "signer", "signed_date"]]
["Read this 8-K and give me JSON with: closing_locations (list of the two places where facilities are being closed, each as 'City, State' exactly as written, in order), press_release_exhibit (object: exhibit_number = the exhibit number furnished, release_date = the date of the press release as written), signer (object:...
[{"closing_locations": ["North Canton, Ohio", "Johnstown, NY"], "press_release_exhibit": {"exhibit_number": "99.1", "release_date": "January 17, 2008"}, "signer": {"name": "John J. Sullivan", "first_title": "Executive Vice President-Finance"}, "signed_date": "January 18, 2008"}]
974
ej3-p00008-e1375a5b09-t1
<|start|>Summarize the header of this securities docket as JSON shaped like {"caseName": "...", "assignedJudge": "...", "dateFiled": "...", "dateTerminated": "...", "defendants": ["...", "..."]}. caseName is the caption as printed; assignedJudge is the judge's name after 'Assigned to:'; dateFiled and dateTerminated are...
extract_json_v3
plain
court_docket
2:13-cv-06808
1
["skeleton_example"]
[["caseName", "assignedJudge", "dateFiled", "dateTerminated", "defendants"]]
["Summarize the header of this securities docket as JSON shaped like {\"caseName\": \"...\", \"assignedJudge\": \"...\", \"dateFiled\": \"...\", \"dateTerminated\": \"...\", \"defendants\": [\"...\", \"...\"]}. caseName is the caption as printed; assignedJudge is the judge's name after 'Assigned to:'; dateFiled and dat...
[{"caseName": "RATZ v. PHOTOMEDEX, INC.", "assignedJudge": "Paul S. Diamond", "dateFiled": "2013-11-22", "dateTerminated": "2015-08-10", "defendants": ["DENNIS M. MCGRATH", "PHOTOMEDEX, INC.", "DOLEV RAFAELI"]}]
2,165
ej3-p00009-d47f1c59d4-t1
<|start|>``` Supplements and Amendments; Whole Agreement This Warrant may be ------------------------------------------- amended or supplemented only by an instrument in writing signed by the parties hereto. This Warrant contains the full understanding of the parties hereto with respect to the subject matter ...
extract_json_v3
plain
edgar_contract
0000944209-01-500335
1
["key_list_with_descriptions"]
[["governingState", "executionDate", "companyName", "signer"]]
["This is the tail end of a stock purchase warrant. Extract: governingState (the state's name whose laws govern, as written); executionDate (the date the warrant was executed, as written); companyName (the company name as printed above the signature line); signer (object with name = the signatory's name as printed, and...
[{"governingState": "Nevada", "executionDate": "January 11, 2001", "companyName": "PCSUPPORT.COM, INC.", "signer": {"name": "David W. Rowat", "title": "Vice President and Chief Financial Officer"}}]
553
ej3-p00009-d47f1c59d4-t2
<|start|>Read the boilerplate and exercise form in this warrant and give me a JSON object with these fields: amendmentMethod (the short phrase describing how the warrant may be amended, starting at 'an instrument'); assignorParty (the defined term for the party allowed to assign the warrant, as written); permittedAssig...
extract_json_v3
plain
edgar_contract
0000944209-01-500335
1
["prose"]
[["amendmentMethod", "assignorParty", "permittedAssignee", "paymentSectionReference", "securityClass", "expirationDate"]]
["Read the boilerplate and exercise form in this warrant and give me a JSON object with these fields: amendmentMethod (the short phrase describing how the warrant may be amended, starting at 'an instrument'); assignorParty (the defined term for the party allowed to assign the warrant, as written); permittedAssignee (th...
[{"amendmentMethod": "an instrument in writing signed by the parties", "assignorParty": "Holder", "permittedAssignee": "accredited investor", "paymentSectionReference": "Section 2", "securityClass": "Common Stock", "expirationDate": null}]
573
ej3-p00009-e10416605c-t1
<|start|><|user|> This is a short 8-K about a transfer agent. Return JSON with: - company: the registrant's name as written in the Item text (with 'Inc.'). - itemNumber: the Item number reported, number only. - effective_date: the date as of which the transfer books were re-opened, as written. - transfer_agent: object ...
extract_json_v3
chat
edgar_8k
0000921895-09-002898
1
["key_list_with_descriptions"]
[["company", "itemNumber", "effective_date", "transfer_agent"]]
["This is a short 8-K about a transfer agent. Return JSON with:\n- company: the registrant's name as written in the Item text (with 'Inc.').\n- itemNumber: the Item number reported, number only.\n- effective_date: the date as of which the transfer books were re-opened, as written.\n- transfer_agent: object with full_na...
[{"company": "Footstar, Inc.", "itemNumber": "8.01", "effective_date": "December 15, 2009", "transfer_agent": {"full_name": "BNY Mellon Shareowner Services", "short_name": "BNY Mellon", "investor_phone": "(800) 851-9677"}}]
876
ej3-p00009-e10416605c-t2
<|start|><<< CURRENT REPORT Pursuant to Section 13 or 15(d) of the Securities Exchange Act of 1934 Date of Report (Date of earliest event reported): December 15, 2009 Footstar, Inc.(Exact name of registrant as specified in its charter)Delaware 1-11681 22-3439443(State or oth...
extract_json_v3
plain
edgar_8k
0000921895-09-002898
1
["skeleton_example"]
[["state_of_incorporation", "commissionFileNumber", "irs_employer_id", "dated"]]
["Fill in the registrant identifiers from the cover page and signature: {\"state_of_incorporation\": \"...\", \"commissionFileNumber\": \"...\", \"irs_employer_id\": \"...\", \"dated\": \"...\"}. state_of_incorporation is the state name; commissionFileNumber is the Commission File Number; irs_employer_id is the IRS Emp...
[{"state_of_incorporation": "Delaware", "commissionFileNumber": "1-11681", "irs_employer_id": "22-3439443", "dated": "December 18, 2009"}]
810
End of preview. Expand in Data Studio

ALEA Exact-Span JSON Extraction

Extraction tasks on real legal and business text. Each row gives a passage and asks for a JSON object with exactly the requested keys. Every value is an exact span of the passage (or null when the passage does not have it).

The set is training data for ALEA Institute language models. It has two configs:

  • diverse (new in v3, the default): document-specific schemas. Each passage, an excerpt of an EDGAR contract exhibit, a Form 8-K or a federal court docket, has its own tasks with 1 to 8 fields (95% of tasks have 3 to 6) chosen for that document, often with nested objects and lists of objects. The rows use 4,252 distinct top-level field names (5,722 counting nested keys).
  • core_spans (the earlier release, unchanged): the six fixed fields dates, defined_terms, dollar_amounts, usc_citations, cfr_citations and fr_citations on EDGAR and Federal Register passages, with answers computed by deterministic code. These are the same 21,709 rows, byte for byte, as release v1 of this repository.
from datasets import load_dataset
diverse = load_dataset("alea-institute/alea-extract-json", "diverse", split="train")
core = load_dataset("alea-institute/alea-extract-json", "core_spans", split="train")

What is in this release

  • diverse: 1,735 rows, 2,032 tasks, 2,286,389 tokens in the ALEA tokenizer, from 1,126 passages.
    • By source: 932 rows from EDGAR contract excerpts, 402 rows from Form 8-K excerpts, 401 rows from court dockets.
    • 297 rows ask two different tasks about the same passage (the second as a follow-up turn); the other 1,438 rows have one task. Each task appears in exactly one row.
    • Formats: 1,125 plain, 317 chat, 293 chat_sys. chat rows use <|user|>, <|assistant|> and <|eot|>; chat_sys rows add a <|system|> line; plain rows are plain text. Every text starts with <|start|> and ends with <|end|> (the encoder strips and checks these).
  • core_spans: 21,709 rows, 12,192,980 tokens, unchanged from v1 (see the core_spans parts of Task and How the answers are made).

Task

diverse

  • Instructions. Each task's instruction was written for its passage by Claude Sonnet, in one of four schema styles: a key list with descriptions (798 tasks), prose naming the keys (642), a JSON skeleton to fill in (310) and a JSON Schema (282). Example: Return JSON with: withholdingMethod (short phrase for how the Required Tax Payments are satisfied, starting 'withholding'), withholdingCap (...), cashAdvanceSection (the Section number cited for paying employment taxes in cash early, as written), .... When an instruction did not name every key (144 tasks), the task's own field list (name and description) is appended to it.
  • Layout. The passage is reproduced verbatim. It is plain or wrapped in ```, <<<, [[[, """ or ~~~ markers. The instruction comes before the passage in 1,074 rows and after it in 661. Some rows have a short opener ("Quick task:", "Hi!") or a system line, from the same lists as core_spans.
  • Answers. One JSON object per task, keys in the instruction's order, indented by 2 or 4 spaces or on one line; some are fenced as ```json or with the passage's own ``` / ~~~ marker. In plain rows the answer follows the instruction and passage directly, with no role markers, as in core_spans.

core_spans

Each row asks for some of the six fixed keys; each value is a list of the exact strings found in the passage, in text order, without duplicates, and an empty list when the passage has none. Rows may have 1 to 5 tasks on different passages (few-shot). Instruction wording comes from hand-written templates and, in 54.6% of rows, template paraphrases written by GLM-5.3-Flash.

How the answers are made

diverse

  • Gold. Fields and values were proposed by Claude Sonnet, independently re-extracted by Claude Opus, and every field where the two disagreed was adjudicated by a fresh Claude Opus call, which chose one answer, accepted both as equivalent, or gave a corrected span (23 fields); it could also reject a field, which would drop the task, but rejected none. Code then dropped tasks with a value that is not an exact span of its passage or with keys other than their fields (22 of 2,492 tasks). Over all 2,492 finalized tasks (before the evaluation-overlap removal below), 1,028 of 10,283 fields (10.0%) needed adjudication.
  • Code checks. Every value is checked by code as an exact substring of its passage (15,493 string values; 173 values are null). Every key, including nested keys, appears literally in the row's instruction text; 22 tasks whose answers had a key (at any level) that the instruction never named were dropped, and 216 tasks were dropped because a multi-word answer value appeared verbatim in their own instruction. Two kinds of multi-word value were exempt from that drop: values that are only a section, item or exhibit reference ("Section 2.8", "Item 9.01"), and values the instruction offers as one of its options ("'Judge' or 'Magistrate Judge'"; an occurrence with "or", "either", "one of" or "/" within 30 characters, unless it follows "e.g.", "for example", "such as" or "like"). Single-word values were not dropped. The top-level keys are exactly the instruction's fields, and 16 tasks with a value that did not have its field's declared shape (string, object, list of strings or list of objects; for example a list of sentences where a list of objects was declared) were dropped, as was 1 task with no fields or only empty values. The shape check uses the declared type, so it can also drop an answer that matched the instruction's wording. Every answer parses back from the row text to its gold, and every row was re-tokenized and decoded back to the same text.
  • No invented values. No language model wrote any passage. The models chose which fields to ask for and which span of the passage answers each field. Every value, including the corrected ones, is text copied from the passage and code-checked as an exact span.

core_spans

Every answer is computed by deterministic code from the passage (regular expressions plus rule checks, one extractor per field); no language model wrote any passage or answer. All 180,285 items are literal spans of their passages. The v1 card's full description of the extractors, drop rules and per-key counts still applies.

Balance and checks

  • Field-name diversity (diverse). 4,252 distinct top-level keys over 8,551 field uses; the 20 most frequent keys cover 17.8% of field uses. Most frequent: caption (201), registrant (148), cause (133), docketNumber (119), signatory (89), commissionFileNumber (75), docket_number (74), signer (73). 2,014 distinct key sets in 2,032 tasks. 70% of tasks have a nested object or a list of objects. Key spelling follows each instruction (camelCase and snake_case both occur, and 84 distinct keys contain spaces, such as "form of payment").
  • Instruction near-duplicates. 0.0% of rows have instruction text (the row minus passage and answers) whose word 5-gram set has a Jaccard similarity above 0.6 with another row's.
  • Nulls. 173 of 15,666 values are null (the passage does not have the item).
  • Overlap with evaluation sets. Passages were compared with private held-out evaluation sets by exact text hash, EDGAR accession number and shared word 13-grams. 92 passages that matched (same accession, or at least 20% of their 13-grams shared) were removed before the build. Below that threshold, overlap remains: of the 1,152 passages kept after the check, 710 share at least one 13-gram with the evaluation references and 576 share at least five (at most 19.8% of a passage's 13-grams). Standard clauses and form language are common in this genre; the remaining overlap was not reviewed further.

Sources

  • EDGAR. Contract exhibits (EX-10) and Form 8-K current reports filed with the U.S. Securities and Exchange Commission, public records. Contract passages are windows of consecutive sections (heading plus body, in document order); 8-K passages are windows of consecutive paragraphs. The text keeps the filing's line breaks, spacing, page markers and OCR errors.
  • Court dockets. Federal district-court civil and criminal dockets from CourtListener (RECAP). The passage is a docket report built from the docket record: the case name, docket number and header fields (assigned judge, dates, cause, nature of suit) as recorded, followed by up to 18 numbered docket entries whose dates and texts are verbatim.
  • Federal Register (core_spans only). Public-domain works of the U.S. government.
  • Names of companies, courts, judges, attorneys and parties in these public records are kept. Passages can contain business contact details (addresses, telephone numbers, notice e-mail addresses) as filed.

Limitations

  • Model-made gold. diverse answers are model judgments (which span answers a field), checked for exactness by code but not by a human. Two models had to agree, or a third call had to rule; this does not guarantee correctness (see Review). The models are from one family (Claude): about 90% of fields rest on Sonnet and Opus agreeing, and the blind reviewers are also Claude Sonnet, so shared blind spots would not be caught.
  • Span boundaries. For phrase-valued fields ("the phrase stating ..."), the start and end of a span can be debatable; the reviewers flagged spans that stop early, a later occurrence used where the first was asked for, and list items left out.
  • Answer cues in instructions. Some instructions name where a span starts ("starting 'withholding'") or quote a term to look for, and 2.3% of answer values (362 of 15,493) appear literally in their own instruction: 268 single words such as party names, 69 section or item references and 25 offered options. Some of these come from example strings that the instruction writer included, which can equal the answer (for example an instruction citing "Section 2.8" as an example). Standard contract wording (Section 409A text, "Common Stock", "Holder") can be guessed without the passage (see Review).
  • Verbatim noise. Values keep the source's doubled spaces, line breaks inside spans and OCR errors ("February l, 2003").
  • Repeated passages. A passage with two tasks appears in two rows unless both tasks are in one row; 609 passages appear in two rows.
  • Domain. diverse covers contracts, 8-Ks and dockets only.
  • core_spans. The v1 limitations still apply: citation over-runs, merged list citations, missed broken-quote definitions, some unextracted citation forms, LLM-paraphrased instructions in 54.6% of rows, and a small overlap with a private evaluation set.

Review

  • Blind review (diverse). Two fresh reviewers (Claude Sonnet) each read 30 different random rows (60 in all), exactly as trained, and checked every value against the passage and instruction. The review and the probe below read the build before the last two code drops (shape mismatches and the one empty task). Rows with a wrong value or wrong structure (HARD): 1 of 60 (1.7%; 95% upper bound 8.9%). Rows with minor issues (SOFT): 6; two of these are incomplete values (a list missing items, a span that stops early), so 3 of 60 rows have a wrong, incomplete or wrongly shaped value. The one HARD row answered a list-of-objects field (one object per arbitrator) with a list of sentences. A code check of declared field shapes, added after the review in response to that finding, dropped 16 tasks; it removed that task (the passage's other task stays as a one-task row) and one task of another reviewed row that the reviewer had rated OK. The other 58 reviewed rows, and 39 of the 40 probe rows, are byte-identical in this release; none of those 58 is HARD (95% upper bound 6.2%, a bound computed after the fix). The SOFT rows: list items left out, a span that stops early, a later occurrence used where the first was asked for, a defendants list not in caption order, a loose attorney-client attribution, and an ambiguous registrant name.
  • Instruction-only probe. A separate reviewer (Claude Opus) saw only the instructions of 40 rows, without the passage, and guessed the answers. Code scored exact matches: 44 of 332 answer values (13.3%) were recovered, mostly stock values and standard wording ("Delaware", "Financial Statements and Exhibits", "specified employee", "Code Section 423", "thirty (30) days") and values that the instruction's wording or examples point to.
  • core_spans. Reviewed for release v1: 0 wrong answers in a final blind check of 80 rows; 1 wrong answer ("$1 millon" gives $1) in a check of 40 of the 161 rows recovered after the narrowed drop rule; and 0 of 226 answer items recovered in an instruction-only probe.

Fields

diverse:

column content
id row id
text the full training text (instructions, passage and answers)
family extract_json_v3
kind chat, chat_sys or plain
source edgar_contract, edgar_8k or court_docket
source_id EDGAR accession number, or the court docket number
n_tasks number of tasks in the row (1 or 2)
schema_styles JSON list, one schema style per task
keys JSON list, the top-level keys of each task
instructions JSON list, the instruction of each task as shown in the row
gold JSON list, one answer object per task
n_tokens tokens in the ALEA tokenizer (alea-v2b-sw32k)

core_spans: id, text, family (extract_json_v2), kind, source (edgar or fr), source_ids, n_shots, keys, gold, n_tokens, as in v1.

License

CC BY 4.0. The underlying texts are public records: filings in the SEC's public EDGAR system, federal court dockets, and Federal Register documents, which are public-domain works of the U.S. government.

Downloads last month
39