Datasets:
id stringlengths 34 34 | text stringlengths 574 11.3k | family stringclasses 1
value | kind stringclasses 5
values | label stringclasses 2
values | source stringclasses 4
values | stratum stringclasses 208
values | gold stringlengths 170 3.44k | presentation stringclasses 3
values | group_id stringlengths 29 39 | n_tokens int64 170 3.53k |
|---|---|---|---|---|---|---|---|---|---|---|
constraint_follow-0002ae8498d83e84 | <|start|><|user|>
For a compliance index, pull the sections out of 7 CFR part 621. Key meanings: number = the section number as printed (e.g. 12.4); subpart = the subpart letter (e.g. B). Return only sections whose heading contains the word “special” (any capitalisation). Return no more than 4 items: the first 4 in the... | constraint_follow | cfr | items | cfr | cfr title 7 | {"items": [["621.44", "E"]], "keys": ["number", "subpart"], "fmt": "table", "filter": {"f": "word", "v": "special"}, "exclude": [], "order": null, "cap": 4, "abstain_clause": false, "n_constraints": 3} | chat | constraint_follow|cfr2005:7:621 | 495 |
constraint_follow-0005c5653505c840 | Below is part of the Federal Register contents of 2021-05-07. Include only entries that start on page 24634 or later. List them in document order. Give at most 16 items; if more qualify, keep the first 16 in the requested order (document order unless another order is requested). Answer as a numbered list; each line is ... | constraint_follow | fr | items | fr | fr 2021 | {"items": [["Commission Decision to Institute a Rescission Proceeding and Rescind the Remedial Orders, to Grant the Motion to Limit Service of the Settlement Agreement, to Deny as Moot the Motion to Terminate, and to Indicate Ruling on Motion to Vacate; Termination of the Rescission Proceeding: Certain Botulinum Toxin ... | plain | constraint_follow|frtoc:2021-05-07 | 1,134 |
constraint_follow-0006f2e7857359aa | <|start|><|user|>
Using only the docket below, extract the docket entries. Sort them by entry number from highest to lowest. If nothing meets these conditions, answer exactly “None stated”. Keep only entries numbered 113 or higher. Return one JSON object that maps the entry number, as an integer (as the key) to the lea... | constraint_follow | docket | items | docket | docket ared | {"items": [["223", "AMENDED"], ["148", "ORDER"], ["147", "ORDER"], ["141", "ORDER"], ["140", "ORDER"], ["115", "ORDER"], ["113", "PROTECTIVE ORDER"]], "keys": ["entry_number", "entry_type"], "fmt": "json_map", "filter": {"f": "num_ge", "v": 113}, "exclude": [], "order": "num_desc", "cap": null, "abstain_clause": true, ... | chat | constraint_follow|dkt:5369725 | 984 |
constraint_follow-0017730b32a1c6ae | <|start|><|user|>
Prepare a list of docket entries from this docket. Answer in YAML: a list of mappings with the keys “filed_on”, “entry” (string values in double quotes). Return no more than 21 item: the first 21 in the requested order (document order if no order is requested). Return only entries filed after August 2... | constraint_follow | docket | items | docket | docket ilnd | {"items": [["2008-09-04", "9"], ["2008-09-04", "10"], ["2008-09-04", "11"], ["2008-09-04", "12"], ["2008-09-04", "13"], ["2008-09-04", "14"], ["2008-09-04", "15"], ["2008-09-04", "16"], ["2008-09-04", "17"], ["2008-09-04", "18"], ["2008-09-04", "19"], ["2008-09-04", "20"], ["2008-09-04", "21"], ["2008-09-04", "22"], ["... | chat | constraint_follow|dkt:4258156 | 2,220 |
constraint_follow-0017cb9c9d82a953 | Build an outline of 42 U.S.C. 16277 from its subsections. Order them alphabetically by heading. Consider only subsections that come after subsection (c). Give a numbered list (1., 2., ...), one item per line: the subsection designation with parentheses, e.g. (b).
Passage:
Title 42, United States Code (2023 release)
§1... | constraint_follow | usc_sub | items | usc | usc title 42 | {"items": [["(d)"]], "keys": ["designation"], "fmt": "numbered", "filter": {"f": "after", "v": "(c)"}, "exclude": [], "order": "alpha", "cap": null, "abstain_clause": false, "n_constraints": 3} | plain | constraint_follow|usc2023:42_16277 | 437 |
constraint_follow-00181d8e92ea235e | Key meanings: document = the entry's title as printed (subject heading and document line joined by ": "); kind = its category heading in title case (e.g. Proposed Rules). Order them alphabetically by title. Consider only entries listed under RULES. Answer as a Markdown table with the columns “document”, “kind”. Make me... | constraint_follow | fr | items | fr | fr 2003 | {"items": [["Bipartisan Campaign Reform Act; implementation: Technical amendments", "Rules"], ["Mining products; testing, evaluation, and approval: Mobile battery-powered machines; plug and receptacle-type connectors; alternate locking devices", "Rules"]], "keys": ["document", "kind"], "fmt": "table", "filter": {"f": "... | plain | constraint_follow|frtoc:2003-01-22 | 706 |
constraint_follow-00197732cc010a14 | <|start|><|system|>
Answer accurately and concisely.<|eot|>
<|user|>
Sort them by entry number from highest to lowest. Include only entries filed after November 17, 2014. Give at most 6 items; if more qualify, keep the first 6 in the requested order (document order unless another order is requested). Answer on one line... | constraint_follow | docket | items | docket | docket wawd | {"items": [["15"], ["14"], ["13"], ["12"], ["11"], ["10"]], "keys": ["doc_no"], "fmt": "comma", "filter": {"f": "after_date", "v": "2014-11-17"}, "exclude": [], "order": "num_desc", "cap": 6, "abstain_clause": false, "n_constraints": 4} | chat | constraint_follow|dkt:4537275 | 1,177 |
constraint_follow-001b057c7dbef8d5 | Query: Here is a solved example on a different document.
Answer in YAML: a list of mappings with the keys “sec”, “subject” (string values in double quotes). Keep only sections whose heading contains the word “accounts” (any capitalisation). Return no more than 2 items: the first 2 in the requested order (document order... | constraint_follow | docket | items | docket | docket nysd | {"items": [["1"], ["2"], ["3"], ["4"], ["5"], ["6"], ["7"], ["8"]], "keys": ["doc_no"], "fmt": "json_objects", "filter": {"f": "on_before", "v": "2015-12-09"}, "exclude": [], "order": null, "cap": 8, "abstain_clause": false, "n_constraints": 3} | qa | constraint_follow|dkt:4356242 | 1,375 |
constraint_follow-001bd88e9d60a181 | Query: Return no more than 1 item: the first 1 in the requested order (document order if no order is requested). title = the entry's title as printed (subject heading and document line joined by ": "). Keep only entries whose title contains the word “prohibited” (any capitalisation). Answer with JSON: an array of objec... | constraint_follow | fr | items | fr | fr 2007 | {"items": [["National Organic Program: National List of Allowed and Prohibited Substances; petition submission guidelines"]], "keys": ["title"], "fmt": "json_objects", "filter": {"f": "word", "v": "prohibited"}, "exclude": [], "order": null, "cap": 1, "abstain_clause": false, "n_constraints": 3} | qa | constraint_follow|frtoc:2007-12-10 | 675 |
constraint_follow-001f364f1d3a6cd4 | <|start|><|system|>
You check your work before answering.<|eot|>
<|user|>
Answer as a numbered list; each line is the entry number, as an integer. Order them by entry number, largest first. Keep only entries whose docket text begins with the word COMPLAINT (in capitals). Index the docket entries in the docket sheet bel... | constraint_follow | docket | items | docket | docket cacd | {"items": [["1"]], "keys": ["entry"], "fmt": "numbered", "filter": {"f": "type", "v": "COMPLAINT"}, "exclude": [], "order": "num_desc", "cap": null, "abstain_clause": false, "n_constraints": 3} | chat | constraint_follow|dkt:4154645 | 849 |
constraint_follow-0027a2c8b2719cfb | <|start|><|user|>
Answer in YAML: a list of mappings with the keys “doc_no” (string values in double quotes). Include only entries whose docket text begins with the word MEMORANDUM (in capitals). doc_no = the entry number, as an integer. Give at most 12 items; if more qualify, keep the first 12 in the requested order (... | constraint_follow | docket | items | docket | docket kywd | {"items": [["30"], ["71"], ["72"], ["73"], ["74"], ["75"], ["76"], ["77"], ["100"], ["115"], ["123"]], "keys": ["doc_no"], "fmt": "yaml", "filter": {"f": "type", "v": "MEMORANDUM"}, "exclude": [], "order": null, "cap": 12, "abstain_clause": true, "n_constraints": 4} | chat | constraint_follow|dkt:4510394 | 1,050 |
constraint_follow-003029a0133bc87b | From the table of contents of 32 CFR part 263 below, list the sections. Answer as a Markdown table with the columns “sec”, “title”. Consider only sections numbered below 263.3 (compare the number after the period). Order them from the last in the document to the first. Key meanings: sec = the section number as printed ... | constraint_follow | cfr | items | cfr | cfr title 32 | {"items": [["263.2", "Applicability."]], "keys": ["sec", "title"], "fmt": "table", "filter": {"f": "secnum_lt", "v": 3}, "exclude": [], "order": "rev", "cap": 1, "abstain_clause": true, "n_constraints": 5} | plain | constraint_follow|cfr2005:32:263 | 302 |
constraint_follow-0033861c9445a601 | Request: Using the part contents below (42 CFR part 90), extract the sections. Answer in YAML: a list of mappings with the keys “cite”, “title” (string values in double quotes). Consider only sections numbered below 90.13 (compare the number after the period). Keep the order in which they appear in the document. Key me... | constraint_follow | cfr | items | cfr | cfr title 42 | {"items": [["90.1", "Purpose and applicability."], ["90.2", "Definitions."], ["90.3", "Procedures for requesting health assessments."], ["90.4", "Contents of requests for health assessments."], ["90.5", "Acting on requests."], ["90.6", "Notification of determination to conduct a health assessment in response to a reque... | qa | constraint_follow|cfr2005:42:90 | 550 |
constraint_follow-00387ee2266a7918 | For a compliance index, pull the sections out of 32 CFR part 1665. Should no item qualify, write only “None stated”. Consider only sections numbered below 1665.1 (compare the number after the period). Key meanings: section_number = the section number as printed (e.g. 12.4); title = the section heading exactly as printe... | constraint_follow | cfr | none | cfr | cfr title 32 | {"items": [], "keys": ["section_number", "title"], "fmt": "json_objects", "filter": {"f": "secnum_lt", "v": 1}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | plain | constraint_follow|cfr2005:32:1665 | 253 |
constraint_follow-00397ed9af2f804b | ### Task
Extract sections from the regulation's table of contents below. Should no item qualify, write only “None stated”. Order them alphabetically by heading. Consider only sections whose heading contains the word “termination” (any capitalisation). Answer as a numbered list; each line is the section number as printe... | constraint_follow | cfr | items | cfr | cfr title 33 | {"items": [["81.17"]], "keys": ["number"], "fmt": "numbered", "filter": {"f": "word", "v": "termination"}, "exclude": [], "order": "alpha", "cap": null, "abstain_clause": true, "n_constraints": 4} | qa | constraint_follow|cfr2005:33:81 | 216 |
constraint_follow-003ac63674893b0d | <|start|><|user|>
Extract sections from the regulation's table of contents below. If no item qualifies, reply exactly “None stated”. Include only sections whose heading contains the word “distsrict” (any capitalisation). Key meanings: section = the section number as printed (e.g. 12.4); title = the section heading exac... | constraint_follow | cfr | items | cfr | cfr title 32 | {"items": [["1657.3", "Distsrict appeal boards."]], "keys": ["section", "title"], "fmt": "yaml", "filter": {"f": "word", "v": "distsrict"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | chat | constraint_follow|cfr2005:32:1657 | 234 |
constraint_follow-00418649a9a2b0b5 | Keep the order in which they appear in the document. Consider only sections whose heading contains the word “construction” (any capitalisation). Key meanings: subpart_letter = the subpart letter (e.g. B); section = the section number as printed (e.g. 12.4); subject = the section heading exactly as printed. Should no it... | constraint_follow | cfr | items | cfr | cfr title 38 | {"items": [["D", "39.19", "General requirements for site selection and construction of veterans' cemeteries."]], "keys": ["subpart_letter", "section", "subject"], "fmt": "json_objects", "filter": {"f": "word", "v": "construction"}, "exclude": [], "order": "doc", "cap": null, "abstain_clause": true, "n_constraints": 4} | plain | constraint_follow|cfr2005:38:39 | 565 |
constraint_follow-00437e25e3515eb8 | From the Federal Register contents of 2003-11-28 (excerpt below), list the entries. Keep the order in which they appear in the document. Include only entries listed under NOTICES. Answer as a numbered list (1., 2., ...), one item per line: the entry's title as printed (subject heading and document line joined by ": "),... | constraint_follow | fr | items | fr | fr 2003 | {"items": [["Competitive impact statements and proposed consent judgments: Univision Communications Inc.", "Notices"], ["Surveys, determinations, etc.: Manufacturing area; annual", "Notices"], ["Surveys, determinations, etc.: Retail trade; annual", "Notices"], ["Grants and cooperative agreements; availability, etc.: As... | plain | constraint_follow|frtoc:2003-11-28 | 926 |
constraint_follow-00443da00d084505 | Go through the section list of 33 CFR part 96. If no item qualifies, reply exactly “None stated”. Return only sections numbered 96.300 or higher (compare the number after the period). Key meanings: cite = the section number as printed (e.g. 12.4); subpart_letter = the subpart letter (e.g. B). Return YAML only: a sequen... | constraint_follow | cfr | items | cfr | cfr title 33 | {"items": [["96.300", "C"], ["96.310", "C"], ["96.320", "C"], ["96.330", "C"], ["96.340", "C"], ["96.350", "C"], ["96.360", "C"], ["96.370", "C"], ["96.380", "C"], ["96.390", "C"], ["96.400", "D"], ["96.410", "D"], ["96.420", "D"], ["96.430", "D"], ["96.440", "D"], ["96.450", "D"], ["96.460", "D"], ["96.470", "D"], ["9... | plain | constraint_follow|cfr2005:33:96 | 1,124 |
constraint_follow-0049dd1d2c30182a | <|start|><|user|>
For a regulatory watch list, go through the Federal Register contents of 2023-06-20. Keep the order in which they appear in the document. Return only entries whose title contains the word “security” (any capitalisation). Key meanings: type = its category heading in title case (e.g. Proposed Rules); pa... | constraint_follow | fr | items | fr | fr 2023 | {"items": [["Notices", "39859", "Agency Information Collection Activities; Proposals, Submissions, and Approvals: Evaluation of Public Housing Agencies Coronavirus Aid, Relief, and Economic Security Act Waivers: Evaluation of Public Housing Agencies Interviews Data Collection"]], "keys": ["type", "page", "subject"], "f... | chat | constraint_follow|frtoc:2023-06-20 | 815 |
constraint_follow-004b3681d69a38f4 | Answer with a single JSON object: keys are the subsection designation with parentheses, e.g. (b), values are the subsection heading exactly as printed. Return only subsections whose heading contains the word “national” (any capitalisation). Sort them alphabetically by heading (A to Z, ignoring case). Below is the text ... | constraint_follow | usc_sub | items | usc | usc title 50 | {"items": [["(b)", "National emergency planning assumptions"]], "keys": ["id", "name"], "fmt": "json_map", "filter": {"f": "word", "v": "national"}, "exclude": [], "order": "alpha", "cap": null, "abstain_clause": false, "n_constraints": 3} | plain | constraint_follow|usc2023:50_98h-5 | 701 |
constraint_follow-0053ab94372fdf8a | <|start|><|user|>
Return only sections numbered 941.974 or higher (compare the number after the period). If no item qualifies, reply exactly “None stated”. Return one JSON object that maps the section number as printed (e.g. 12.4) (as the key) to the section heading exactly as printed (as the value). Please list the se... | constraint_follow | cfr | none | cfr | cfr title 30 | {"items": [], "keys": ["section_number", "title"], "fmt": "json_map", "filter": {"f": "secnum_ge", "v": 974}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | chat | constraint_follow|cfr2005:30:941 | 703 |
constraint_follow-0054292079fe6f8f | entry = the entry's title as printed (subject heading and document line joined by ": "). Keep the order in which they appear in the document. List only entries listed under PROCLAMATIONS. If no item qualifies, reply exactly “None stated”. Answer as a Markdown table with the columns “entry”. From the Federal Register co... | constraint_follow | fr | items | fr | fr 2001 | {"items": [["Terrorist attacks; declaration of national emergency (Proc. 7463)"]], "keys": ["entry"], "fmt": "table", "filter": {"f": "category", "v": "Proclamations"}, "exclude": [], "order": "doc", "cap": null, "abstain_clause": true, "n_constraints": 4} | plain | constraint_follow|frtoc:2001-09-18 | 716 |
constraint_follow-0058115c7b67646f | <|start|><|system|>
You are a precise assistant for text and documents.<|eot|>
<|user|>
Below is part of the Federal Register contents of 2001-04-18. Return a JSON array of strings (one value per item: the entry's title as printed (subject heading and document line joined by ": ")). Do not include the entry titled “App... | constraint_follow | fr | items | fr | fr 2001 | {"items": [["Applications, hearings, determinations, etc.: Duke Cogema Stone & Webster"]], "keys": ["subject"], "fmt": "json_strings", "filter": {"f": "word", "v": "hearings"}, "exclude": ["the entry titled “Applications, hearings, determinations, etc.: Southern Nuclear Operating Co.”", "the entry titled “Applications,... | chat | constraint_follow|frtoc:2001-04-18 | 754 |
constraint_follow-005b51c29b378c21 | Instruction: Keep only subsections that come after subsection (c). List them in document order. Answer as a numbered list (1., 2., ...), one item per line: the subsection designation with parentheses, e.g. (b), then the subsection heading exactly as printed, joined by “ — ”. Below is the text of 7 U.S.C. 2009cc-5.
Pas... | constraint_follow | usc_sub | items | usc | usc title 7 | {"items": [["(d)", "Subrogation and ownership rights"], ["(e)", "Management and administration"]], "keys": ["designation", "title"], "fmt": "numbered", "filter": {"f": "after", "v": "(c)"}, "exclude": [], "order": "doc", "cap": null, "abstain_clause": false, "n_constraints": 3} | qa | constraint_follow|usc2023:7_2009cc-5 | 644 |
constraint_follow-0062b5c87045930d | ### Task
List only subsections whose heading contains the word “evidentiary” (any capitalisation). Key meanings: designation = the subsection designation with parentheses, e.g. (b); name = the subsection heading exactly as printed. Should no item qualify, write only “None stated”. Sort them alphabetically by heading (A... | constraint_follow | usc_sub | items | usc | usc title 20 | {"items": [["(e)", "Review and evidentiary functions"]], "keys": ["designation", "name"], "fmt": "table", "filter": {"f": "word", "v": "evidentiary"}, "exclude": [], "order": "alpha", "cap": 1, "abstain_clause": true, "n_constraints": 5} | qa | constraint_follow|usc2023:20_1234 | 869 |
constraint_follow-00662d12d8859731 | <|start|><|user|>
Answer with JSON: an array of objects, keys “designation”, “heading”. List them in reverse document order (last one first). Do not include subsection (g) and subsection (d). Key meanings: designation = the subsection designation with parentheses, e.g. (b); heading = the subsection heading exactly as p... | constraint_follow | usc_sub | items | usc | usc title 42 | {"items": [["(f)", "Coordination"], ["(e)", "Meetings"]], "keys": ["designation", "heading"], "fmt": "json_objects", "filter": {"f": "after", "v": "(c)"}, "exclude": ["subsection (g)", "subsection (d)"], "order": "rev", "cap": null, "abstain_clause": false, "n_constraints": 4} | chat | constraint_follow|usc2023:42_300hh-10b | 1,047 |
constraint_follow-006830a07ffcdb48 | Read 26 U.S.C. 280C and extract its headed subsections. Output only a JSON list of strings, each being the subsection designation with parentheses, e.g. (b). Sort them alphabetically by heading (A to Z, ignoring case). Include only subsections whose heading contains the word “testing” (any capitalisation). If nothing m... | constraint_follow | usc_sub | items | usc | usc title 26 | {"items": [["(b)"]], "keys": ["designation"], "fmt": "json_strings", "filter": {"f": "word", "v": "testing"}, "exclude": [], "order": "alpha", "cap": null, "abstain_clause": true, "n_constraints": 4} | plain | constraint_follow|usc2023:26_280C | 1,077 |
constraint_follow-007db2b837dbf36f | <|start|><|user|>
For a case chronology, go through the docket below. Return only entries filed on or before September 26, 2016. Do not include entry 7. Answer as a numbered list; each line is the entry number, as an integer.
"""
Docket entries (court: okwd; docket 5253048)
Date Filed | # | Docket Text
08/30/2016 | 1 ... | constraint_follow | docket | items | docket | docket okwd | {"items": [["1"], ["2"], ["3"], ["4"], ["5"], ["6"]], "keys": ["entry_no"], "fmt": "numbered", "filter": {"f": "on_before", "v": "2016-09-26"}, "exclude": ["entry 7"], "order": null, "cap": null, "abstain_clause": false, "n_constraints": 3} | chat | constraint_follow|dkt:5253048 | 943 |
constraint_follow-007f62c0493f0e00 | Request: Using the part contents below (20 CFR part 228), extract the sections. Produce a Markdown table whose header row is “number”. number = the section number as printed (e.g. 12.4). Include only sections in Subpart B. Do not include § 228.11 and § 228.13. List them in document order.
Text:
Code of Federal Regulat... | constraint_follow | cfr | items | cfr | cfr title 20 | {"items": [["228.10"], ["228.12"], ["228.14"], ["228.15"], ["228.16"], ["228.17"], ["228.18"], ["228.19"], ["228.20"], ["228.21"], ["228.22"], ["228.23"], ["228.40"]], "keys": ["number"], "fmt": "table", "filter": {"f": "subpart", "v": "B"}, "exclude": ["§ 228.11", "§ 228.13"], "order": "doc", "cap": null, "abstain_cla... | qa | constraint_follow|cfr2005:20:228 | 683 |
constraint_follow-0082b6961225fde8 | Give a numbered list; each line is the entry number, as an integer, then the filing date in YYYY-MM-DD form, separated by “ — ”. If no item qualifies, reply exactly “None stated”. Consider only entries whose docket text begins with the word SEALED (in capitals). Read the docket entries below.
---
Docket entries (court... | constraint_follow | docket | items | docket | docket txed | {"items": [["15", "2015-03-20"]], "keys": ["entry", "filed"], "fmt": "numbered", "filter": {"f": "type", "v": "SEALED"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | plain | constraint_follow|dkt:4387363 | 1,107 |
constraint_follow-0089c98d8495857a | <|start|><|user|>
Read the contents listing of 7 CFR part 2901. Include only sections numbered below 2901.7 (compare the number after the period). Omit § 2901.3. Key meanings: section = the section number as printed (e.g. 12.4); heading = the section heading exactly as printed. Return no more than 5 items: the first 5 ... | constraint_follow | cfr | items | cfr | cfr title 7 | {"items": [["2901.1", "Purpose and scope."], ["2901.2", "Definitions."], ["2901.4", "Interpretations."], ["2901.5", "Modifications and rescissions."], ["2901.6", "Exceptions and exemptions."]], "keys": ["section", "heading"], "fmt": "table", "filter": {"f": "secnum_lt", "v": 7}, "exclude": ["§ 2901.3"], "order": null, ... | chat | constraint_follow|cfr2005:7:2901 | 341 |
constraint_follow-00927bde35687fac | Query: Answer as a numbered list (1., 2., ...), one item per line: the agency heading it is listed under, then the entry's title as printed (subject heading and document line joined by ": "), then the first page number, as an integer, joined by “ — ”. Return only entries listed under General Services Administration. Ca... | constraint_follow | fr | items | fr | fr 2013 | {"items": [["General Services Administration", "Acquisition Regulations: Electronic Contracting Initiative", "31879"]], "keys": ["issuing_agency", "title", "pg"], "fmt": "numbered", "filter": {"f": "agency", "v": "General Services Administration"}, "exclude": [], "order": null, "cap": 2, "abstain_clause": false, "n_con... | qa | constraint_follow|frtoc:2013-05-28 | 712 |
constraint_follow-009b17c6e66dcbb9 | Below is the contents page of 30 CFR part 717. Answer with JSON: an array of objects, keys “section”. Return no more than 1 item: the first 1 in the requested order (document order if no order is requested). List only sections whose heading contains the word “obligations” (any capitalisation). section = the section num... | constraint_follow | cfr | items | cfr | cfr title 30 | {"items": [["717.11"]], "keys": ["section"], "fmt": "json_objects", "filter": {"f": "word", "v": "obligations"}, "exclude": [], "order": null, "cap": 1, "abstain_clause": false, "n_constraints": 3} | plain | constraint_follow|cfr2005:30:717 | 290 |
constraint_follow-009d1768999e33d6 | <|start|><|system|>
You check your work before answering.<|eot|>
<|user|>
Read the Federal Register contents excerpt below. Answer with JSON: an array of objects, keys “subject”, “kind”, “first_page”. Should no item qualify, write only “None stated”. Keep only entries listed under National Endowment for the Humanities.... | constraint_follow | fr | items | fr | fr 2016 | {"items": [["Meetings: Humanities Panel", "Notices", "63499"]], "keys": ["subject", "kind", "first_page"], "fmt": "json_objects", "filter": {"f": "agency", "v": "National Endowment for the Humanities"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | chat | constraint_follow|frtoc:2016-09-15 | 885 |
constraint_follow-009f74b16a3cc38b | Task: Go through the section list of 12 CFR part 619. Return a JSON array of objects with exactly the keys “sec”, “title”, in that order. List only sections whose heading contains the word “cooperatives” (any capitalisation). List them in document order. Key meanings: sec = the section number as printed (e.g. 12.4); ti... | constraint_follow | cfr | items | cfr | cfr title 12 | {"items": [["619.9060", "Bank for cooperatives."]], "keys": ["sec", "title"], "fmt": "json_objects", "filter": {"f": "word", "v": "cooperatives"}, "exclude": [], "order": "doc", "cap": null, "abstain_clause": false, "n_constraints": 3} | qa | constraint_follow|cfr2005:12:619 | 486 |
constraint_follow-00a1a50ad9ed35ed | Return a JSON array of integers (one value per item: the entry number, as an integer). Consider only entries filed on or before December 24, 2015. Order them by entry number, largest first. Index the docket entries in the docket sheet below.
<text>
Docket entries (court: prd; docket 4539808)
Date Filed | # | Docket Te... | constraint_follow | docket | items | docket | docket prd | {"items": [["8"], ["7"], ["6"], ["5"], ["4"], ["3"], ["2"], ["1"]], "keys": ["entry_no"], "fmt": "json_strings", "filter": {"f": "on_before", "v": "2015-12-24"}, "exclude": [], "order": "num_desc", "cap": null, "abstain_clause": false, "n_constraints": 3} | plain | constraint_follow|dkt:4539808 | 1,024 |
constraint_follow-00a310258aced8a5 | <|start|><|user|>
Below is the contents page of 12 CFR part 573. Keep the order in which they appear in the document. Should no item qualify, write only “None stated”. Include only sections whose heading contains the word “transactions” (any capitalisation). Cap the list at 1 item, taken from the top of the requested o... | constraint_follow | cfr | items | cfr | cfr title 12 | {"items": [["573.14"]], "keys": ["section"], "fmt": "numbered", "filter": {"f": "word", "v": "transactions"}, "exclude": [], "order": "doc", "cap": 1, "abstain_clause": true, "n_constraints": 5} | chat | constraint_follow|cfr2005:12:573 | 457 |
constraint_follow-00a4d3c9c22e0150 | Task: Omit entry 49. entry_number = the entry number, as an integer. Include only entries numbered 49 or higher. Keep the order in which they appear in the document. Return a JSON array of objects with exactly the keys “entry_number”, in that order. Prepare a list of docket entries from this docket.
---
Docket entries... | constraint_follow | docket | items | docket | docket wiwd | {"items": [["60"], ["70"]], "keys": ["entry_number"], "fmt": "json_objects", "filter": {"f": "num_ge", "v": 49}, "exclude": ["entry 49"], "order": "doc", "cap": null, "abstain_clause": false, "n_constraints": 4} | qa | constraint_follow|dkt:5606728 | 983 |
constraint_follow-00b089bddf49a86f | <|start|><|user|>
Extract sections from the regulation's table of contents below. Consider only sections numbered below 222.9 (compare the number after the period). Key meanings: section_number = the section number as printed (e.g. 12.4); subpart = the subpart letter (e.g. B). Do not include § 222.7 and § 222.8. Order ... | constraint_follow | cfr | items | cfr | cfr title 36 | {"items": [["222.6", "A"], ["222.4", "A"], ["222.3", "A"], ["222.2", "A"], ["222.1", "A"]], "keys": ["section_number", "subpart"], "fmt": "json_objects", "filter": {"f": "secnum_lt", "v": 9}, "exclude": ["§ 222.7", "§ 222.8"], "order": "rev", "cap": null, "abstain_clause": false, "n_constraints": 4} | chat | constraint_follow|cfr2005:36:222 | 796 |
constraint_follow-00b17bedbc52c666 | Request: Using only the docket below, extract the docket entries. Order them from the last in the document to the first. Exclude entry 64 and entry 67. Keep only entries numbered 31 or higher. Return one JSON object that maps the entry number, as an integer (as the key) to the filing date in YYYY-MM-DD form (as the val... | constraint_follow | docket | items | docket | docket mied | {"items": [["79", "2019-06-26"], ["68", "2014-09-26"], ["62", "2014-08-14"], ["43", "2013-07-01"], ["40", "2013-05-29"], ["31", "2012-10-10"]], "keys": ["doc_no", "date"], "fmt": "json_map", "filter": {"f": "num_ge", "v": 31}, "exclude": ["entry 64", "entry 67"], "order": "rev", "cap": null, "abstain_clause": false, "n... | qa | constraint_follow|dkt:5239767 | 716 |
constraint_follow-00b1e7a06d1ab25e | <|start|><|user|>
Produce a Markdown table whose header row is “entry_number”, “date”. Key meanings: entry_number = the entry number, as an integer; date = the filing date in YYYY-MM-DD form. Return only entries filed after November 24, 2015. Order them by entry number, largest first. Using only the docket below, extra... | constraint_follow | docket | items | docket | docket nysd | {"items": [["19", "2016-02-09"], ["18", "2016-01-21"], ["17", "2016-01-20"], ["16", "2016-01-20"], ["15", "2016-01-20"]], "keys": ["entry_number", "date"], "fmt": "table", "filter": {"f": "after_date", "v": "2015-11-24"}, "exclude": [], "order": "num_desc", "cap": null, "abstain_clause": false, "n_constraints": 3} | chat | constraint_follow|dkt:4356028 | 1,875 |
constraint_follow-00b50b24d4649c36 | Include only subsections whose heading contains the word “proposals” (any capitalisation). If nothing meets these conditions, answer exactly “None stated”. Give a numbered list; each line is the subsection designation with parentheses, e.g. (b), then the subsection heading exactly as printed, separated by “ — ”. Build ... | constraint_follow | usc_sub | items | usc | usc title 22 | {"items": [["(a)", "Congressional authorization; proposals"]], "keys": ["id", "heading"], "fmt": "numbered", "filter": {"f": "word", "v": "proposals"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | plain | constraint_follow|usc2023:22_2803 | 816 |
constraint_follow-00b63dee9f2bc85d | Prepare a list of docket entries from this docket. Include only entries whose docket text begins with the word WAIVER (in capitals). Give at most 3 items; if more qualify, keep the first 3 in the requested order (document order unless another order is requested). Keep the order in which they appear in the document. Ans... | constraint_follow | docket | items | docket | docket laed | {"items": [["6", "2017-03-16"]], "keys": ["entry_number", "filed_on"], "fmt": "json_map", "filter": {"f": "type", "v": "WAIVER"}, "exclude": [], "order": "doc", "cap": 3, "abstain_clause": false, "n_constraints": 4} | plain | constraint_follow|dkt:6147531 | 2,076 |
constraint_follow-00bbe9458e6c8a2c | Below is part of the Federal Register contents of 2021-04-19. Answer in YAML: a list of mappings with the keys “title” (string values in double quotes). Cap the list at 1 item, taken from the top of the requested order (or of the document order if none is requested). Consider only entries whose title contains the word ... | constraint_follow | fr | items | fr | fr 2021 | {"items": [["Agency Information Collection Activities; Proposals, Submissions, and Approvals: The Maternal, Infant, and Early Childhood Home Visiting Program Performance Measurement Information System"]], "keys": ["title"], "fmt": "yaml", "filter": {"f": "word", "v": "measurement"}, "exclude": [], "order": null, "cap":... | plain | constraint_follow|frtoc:2021-04-19 | 848 |
constraint_follow-00bd3aca48b64c3d | <|start|><|system|>
You check your work before answering.<|eot|>
<|user|>
Example (another document):
From the table of contents of 28 CFR part 40 below, list the sections. Output only a JSON list of strings, each being the section number as printed (e.g. 12.4). Leave out § 40.9 even if it qualifies. Return only sectio... | constraint_follow | usc_sub | items | usc | usc title 15 | {"items": [["(f)", "Recipients' duty to furnish information"]], "keys": ["subsection", "name"], "fmt": "table", "filter": {"f": "word", "v": "duty"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | chat | constraint_follow|usc2023:15_649b | 1,349 |
constraint_follow-00caacbef436c5a6 | [Task]
Read the contents listing of 38 CFR part 12. Produce a Markdown table whose header row is “number”. Keep only sections numbered 12.21 or higher (compare the number after the period). If nothing meets these conditions, answer exactly “None stated”. Order them from the last in the document to the first. Cap the li... | constraint_follow | cfr | items | cfr | cfr title 38 | {"items": [["12.24"], ["12.23"]], "keys": ["number"], "fmt": "table", "filter": {"f": "secnum_ge", "v": 21}, "exclude": [], "order": "rev", "cap": 2, "abstain_clause": true, "n_constraints": 5} | qa | constraint_follow|cfr2005:38:12 | 641 |
constraint_follow-00ced9146620c09d | Below is the contents page of 42 CFR part 416. Give at most 23 items; if more qualify, keep the first 23 in the requested order (document order unless another order is requested). List only sections numbered 416.40 or higher (compare the number after the period). If nothing meets these conditions, answer exactly “None ... | constraint_follow | cfr | items | cfr | cfr title 42 | {"items": [["416.40", "Condition for coverage—Compliance with State licensure law."], ["416.41", "Condition for coverage—Governing body and management."], ["416.42", "Condition for coverage—Surgical services."], ["416.43", "Condition for coverage—Evaluation of quality."], ["416.44", "Condition for coverage—Environment.... | plain | constraint_follow|cfr2005:42:416 | 1,023 |
constraint_follow-00cf430a40fed9d4 | Output only a JSON array; each element is an object with the keys “doc_no”, “filed” (in this order). List them in document order. Return no more than 1 item: the first 1 in the requested order (document order if no order is requested). Include only entries whose docket text begins with the word MOTION (in capitals). Ke... | constraint_follow | docket | items | docket | docket dcd | {"items": [["6", "2017-01-06"]], "keys": ["doc_no", "filed"], "fmt": "json_objects", "filter": {"f": "type", "v": "MOTION"}, "exclude": [], "order": "doc", "cap": 1, "abstain_clause": false, "n_constraints": 4} | plain | constraint_follow|dkt:4498951 | 1,828 |
constraint_follow-00d238e5adf422d9 | Give a numbered list (1., 2., ...), one item per line: the subsection designation with parentheses, e.g. (b). If nothing meets these conditions, answer exactly “None stated”. Return only subsections whose heading contains the word “completions” (any capitalisation). Below is the text of 7 U.S.C. 5921.
```
Title 7, Uni... | constraint_follow | usc_sub | none | usc | usc title 7 | {"items": [], "keys": ["subsection"], "fmt": "numbered", "filter": {"f": "word", "v": "completions"}, "exclude": [], "order": null, "cap": null, "abstain_clause": true, "n_constraints": 3} | plain | constraint_follow|usc2023:7_5921 | 550 |
ALEA constraint_follow: multi-constraint instruction following over real legal and government documents
This dataset has 14,936 rows, 13.40M tokens under the alea-v2b-sw32k tokenizer, drawn from 9,089 source documents. Each row pairs one real document with an instruction. The instruction asks for items taken from the document's own structure and sets 3 to 5 constraints that must all hold at once and can each be checked:
- Output format. One of seven formats, using key names that change from row to row:
- JSON array of objects;
- JSON array of strings or of integers;
- JSON object map;
- YAML;
- Markdown table;
- numbered list;
- comma-separated list.
- Filter. An inclusion rule, for example: only sections in Subpart B; only terms from P to Z; only entries filed after a given date; only entries listed under a given agency; only headings containing a given word.
- Exclusion. Named items to leave out (22% of rows).
- Ordering. Document, reverse, alphabetical or numeric descending order (76%).
- Cap. A maximum number of items (40%).
- Abstention clause. "If no item qualifies, reply exactly “None stated”" (36%). In 6.2% of rows nothing qualifies.
Row mix by constraint count: 3 constraints 41%, 4 constraints 44%, 5 constraints 16%.
About 7.8% of rows open with a worked example on a document from a different source. The example always uses a different format and different key names from the target, so copying it gives a wrong answer.
Sources
All texts are public-record U.S. government and court material from the KL3M data collection. They were read from local copies; nothing was fetched for this dataset.
| source | document | items | docs | share |
|---|---|---|---|---|
| Code of Federal Regulations, annual edition 2005 (GPO) | one part's table of contents | section number, heading, subpart | 4,474 | 29.9% |
| Federal Register daily issues 2000-2023 | an excerpt of the issue's table of contents (4+ agencies, 12-32 entries) | agency, category, title, first page | 4,474 | 29.9% |
| Federal court dockets (CourtListener / RECAP docket entries) | one docket sheet (10-30 entries) | entry number, filing date, leading capitalised words | 4,064 | 27.2% |
| United States Code (2023 release, Office of the Law Revision Counsel) | one section | headed subsections, or defined terms | 1,924 | 12.9% |
No court, CFR title, U.S. Code title or Federal Register year exceeds 5% of rows; the largest is one district court at 4.9%. Dockets containing a private e-mail address or phone number were excluded. Party, judge and attorney names are public record and appear as filed.
How the answers were made
No language model was used. The instructions come from more than 8 hand-written wordings per element, and code built every answer:
- Items. Items are read from the document structure: CFR contents XML, Federal Register contents XML, U.S. Code heading and definition elements, and docket entry fields.
- Gold list. The filter, exclusion, ordering and cap are applied in code, and the response is rendered from that list.
- Parse check. Every response was parsed back with a real parser (JSON, YAML, table and list parsers) and had to equal the gold list exactly. This check ran when rows were built and again on the final files.
- Values. Every output value appears in the document as printed. The only exception is dates, which are asked for in YYYY-MM-DD form and checked against the printed MM/DD/YYYY form.
- Excluded documents. A definitions section is skipped if any paragraph defines more than one term, or if any numbered paragraph at the defining level does not define a term.
- Exclusions. Excluded items are always named unambiguously.
Decontamination and checks
Rows were screened against ALEA evaluation and probe sets and against held-out rows of other ALEA task families. Only real passage overlap was dropped.
Cue checks used a question-only bag-of-words classifier, scored with 5-fold grouped cross-validation, to predict whether a row's answer is a list or "None stated". The classifier stays at or below the majority-class prior for every source:
| source | classifier accuracy | majority-class prior |
|---|---|---|
| cfr | 90.2% | 93.4% |
| fr | 90.3% | 93.5% |
| docket | 92.6% | 95.1% |
| usc subsections | 88.1% | 92.3% |
Abstention rows whose instruction alone gave the answer away were removed. Filter values that match nothing are drawn from the vocabulary of other documents.
Review
Independent reviewers checked rendered rows against the documents in blind rounds of 60 rows, each on fresh documents:
- Round 1. Found 5 errors:
- integers requested as "strings";
- definitions whose designation or coverage was wrong;
- an ambiguous exclusion prefix.
- Round 2. Found 1 error: a second term defined in the same paragraph.
- Fixes. All of these were fixed in the generator and the data was rebuilt.
- Round 3. On the rebuilt data, 0 of 60 rows were wrong. Three rows were flagged for awkward wording only: "in this order" is written even when there is a single key.
In addition, the builder read 30 rows against their documents.
Format
Parquet, one train split. The columns are:
id,text(the full training document),family,kind(cfr / fr / docket / usc_sub / usc_def),source,stratum;group_id(the source document),presentation(plain / qa / chat);gold(JSON: items, keys, format, filter, exclusions, order, cap, constraint count);label(items / none),n_tokens.
Each text is at most 4,096 tokens.
License
The dataset compilation is released under CC BY 4.0. The underlying government texts are public domain, and the court docket entries are public record.
- Downloads last month
- 66