Dataset Viewer
Auto-converted to Parquet Duplicate
task_id
int32
18
100
area
stringclasses
6 values
prompt
stringclasses
6 values
golden_answer
stringclasses
6 values
associated_rubrics
int32
29
43
rubrics
listlengths
29
43
task_details
stringclasses
6 values
reference_materials
stringclasses
6 values
realistic_explanation
stringclasses
6 values
difficulty_explanation
stringclasses
6 values
peer_review_1_overall_quality_label
stringclasses
2 values
peer_review_1_overall_quality_description
stringclasses
6 values
peer_review_1_representativeness_label
stringclasses
3 values
peer_review_1_difficulty_label
stringclasses
3 values
peer_review_1_difficulty_description
stringclasses
6 values
peer_review_1_task_specification_quality_label
stringclasses
2 values
peer_review_2_overall_quality_label
stringclasses
2 values
peer_review_2_overall_quality_description
stringclasses
6 values
peer_review_2_representativeness_label
stringclasses
3 values
peer_review_2_difficulty_label
stringclasses
3 values
peer_review_2_difficulty_description
stringclasses
6 values
peer_review_2_task_specification_quality_label
stringclasses
1 value
93
Corporate & M&A
You are a corporate lawyer in England and Wales. An England and Wales incorporated private limited company requires assistance as follows: The Client has one majority shareholder (75% holding), and 3 other investors, all party to a shareholders agreement. One investor (10% shareholder, not a director or employee) has b...
**Background** We have been asked to set out some advice to the founder of a company incorporated in England and Wales (the "**Company**"). We understand that the Company has one majority shareholder (the "**Founder**") holding 75% fully paid up shares (same class as all shareholders) and three minority investors, all ...
30
[ { "rubric_category": "Structure & Style Rubric", "score_option": 5, "criterion": "Does the answer use defined terms consistently throughout?", "justification": "Legal outputs should match legal documentation style and consistent use of defined terms is an important marker of professionalism and lega...
Share buybacks are nuanced because there is a legal process in the UK, but sometimes the commercial realities are rather nuanced. Often corporate lawyers deal with shareholder breaches and company disputes. However this hinges on the lawyers and the finance professional working together to understand the financial and ...
https://www.legislation.gov.uk/ukpga/2006/46/contents
This is a common scenario - breaching SH, company wanting to act but the financials creating a barrier.
I think lawyers and I suspect perhaps AI may get a bit lost with the right path forward. The numbers may need changing - the point is that it might not be possible to do the share buyback, so perhaps it falls on the Founder to somehow do this (which again is not so straightforward). The AI might force a solution, but t...
4 - Above Standards
I would suggesting including options that are unattractive (for completeness) and also those that won't work (to show they have been considered). More descriptive subheadings would assist with rapid absorption of the information
5 - Core Activity
5 - Very Hard
Requires consideration of: - unfair prejudice - vague and nebulous claim to bring - buybacks - fiddly, multiple options, funding - purchase by founder: - impact on other shareholders - funding - distributions - directors duties
Well-Specified
4 - Above Standards
The memo is drafted professionally and the advice provided were clear, concise, and most importantlly commercially smart and viable. This is usually very appreciated by clients who are looking for legal solutions that will also make the most sense to them commercially after evaluating the circumstances and challenges a...
4 - Typical
3 - Moderate
The complexity of this case is less in the law, but slightly more on the accounting side of the shares arrangement and calculation. The legal provisions and analysis are relatively straightforward.
Well-Specified
68
Commercial Litigation
You are an associate attorney at a large law firm, who has been assigned to a new matter involving a contractual dispute between two AI companies. You represent Brightmind AI ("Brightmind"), who is the defendant in a civil litigation matter brought by plaintiff Orion AI ("Orion"). The two companies previously entered i...
**I. Introduction** This memorandum discusses the legal issues raised by the document subpoenas our client, Brightmind AI ("Brightmind") has issued to third parties XYZ, Inc. ("XYZ") and ABC AI ("ABC") in the _Orion v. Brightmind _litigation matter. As you know, XYZ is an IT vendor used by plaintiff Orion AI ("Orion"...
34
[ { "rubric_category": "Structure & Style Rubric", "score_option": 3, "criterion": "Is the response organized into clearly labeled, distinct sections?", "justification": "Effective legal writing should be organized into distinct sections, with each separate legal issue being analyzed separately." },...
In general, a third party not directly involved in litigation has no duty to _preserve_ relevant evidence; rather, their obligation only extends to producing such documents when in receipt of a subpoena under Federal Rule of Civil Procedure 45. However, there are certain elements which can create such an obligation, su...
<https://www.wcslaw.com/accolades/does-a-third-party-have-a-duty-to-preserve-esi/> <https://www.butlersnow.com/news-and-events/wait-why-am-i-receiving-this-practice-pointers-on-third-party-responses-to-litigation-preservation-demands> <https://www.lexology.com/library/detail.aspx?g=a00248bb-0651-4bc1-b8eb-c3b6113a846...
This is a very realistic task - in large-scale civil litigation matters, parties often must issue document subpoenas to third parties under Federal Rule of Civil Procedure 45. This prompt puts a bit of a different spin on the issue, as it involves a party first sending an initial _preservation notice_ to the third part...
I would say this is somewhat tricky - the issue of third party _preservation _obligations (as opposed to production obligations pursuant to a subpoena) is somewhat uncommon. I am introducing two different third parties in this prompt, each with different relationships to the defendant, and each with a different set of ...
4 - Above Standards
This is a realistic litigation discovery scenario involving subpoenas, non-party discovery obligations, preservation duties, and proportionality. The prompt mirrors the type of analytical work a lawyer would perform when evaluating subpoena strategy.
4 - Typical
3 - Moderate
This prompt requires good legal judgment and familiarity with civil discovery rules, but it does not require particularly deep or novel legal research. A mid-level or junior litigation attorney could complete it without a lot of difficulty
Well-Specified
5 - Exemplary
Very well written professional work. I enjoyed reading it.
5 - Core Activity
4 - Hard
Requires deep expertise on the subject.
Well-Specified
100
Contract Law
You are the General Counsel of Acme Corporation, a non-profit organization based in Northern Virginia, with annual revenue of Five Million Dollars. Your organization is recognized by the Internal Revenue Service as a 501(c)(3) tax exempt organization. The Board of Directors has told you that the current CEO is retirin...
To: Board of Directors From: General Counsel Date: February 27, 2026 Subject: Draft Employment Agreement; Potential Legal Issues with Proposed Agreement **A. Introduction: ** You have asked for a draft employment agreement for the hiring of the new CEO. That draft is attached. You have also asked for this memo ident...
29
[ { "rubric_category": "Structure & Style Rubric", "score_option": 5, "criterion": "Does the memo to the Board include identification of key issues, analysis of those issues, citation of sources, and assumptions made?", "justification": null }, { "rubric_category": "Structure & Style Rubric", ...
I tried to create a prompt that would test the model's ability to identify and address legal issues that were not obvious from the prompt; here, that issue is the IRS restriction on excessive compensation for the executives of tax exempt organizations. I wanted to see if the model would identify that issue, including t...
The primary source is IRC § 4958, and the "safe harbor" provisions of the IRS regulations promulgated thereunder.
I would say a 4, as it is a very real issue facing tax exempt organizations, with serious consequences for the directors that approve compensation deemed excessive.
It's a 4, not so much for the drafting itself, but because of the issues that must be identified and incorporated into that drafting.
4 - Above Standards
This is a competent, thoughtful first-draft work that a Board could act on with clarifications, but falls short of elite GC standards due to client-accessibility tweaks, deeper explanations/proposals on open terms, stronger risk wording, and minor factual/drafting cleanups.
4 - Typical
3 - Moderate
The task requires integrating specialized knowledge across three distinct areas—nonprofit tax compliance, Virginia employment/non-compete law, and executive contract drafting.
Well-Specified
5 - Exemplary
It succeeds by forcing the model to solve a "Double Trap" involving federal tax law and state employment rules. The prompt places the writer in a high-stakes role as General Counsel. The respondent must warn the Board of personal financial risks under IRS Section 4958. The Gold Answer provides a clear summary of these ...
5 - Core Activity
5 - Very Hard
It requires deep knowledge in two separate areas of law. A writer must navigate the IRS "safe harbor" rules while also applying Virginia’s strict three-part test for non-compete clauses. The task is hard because it sets a trap. The Board wants a five-year, three-country ban. In Virginia, this is legally void. To pass, ...
Well-Specified
18
Employment & Labor
A senior marketing manager, Sarah, is employed by a multinational company based in London. Following her maternity leave, she requested hybrid working (working from home/remotely three days a week). The employer rejected her request on the basis of ‘team cohesion’ and the need for teams to work together in person. Howe...
**Legal Memorandum** To: Sarah (Client) **Subject: Legal Analysis of Employment Dispute Claim ** [ ] **1. EXECUTIVE SUMMARY** Based on the facts provided, Sarah will have strong grounds to pursue her claims against her former employer in the Employment Tribunal. Her strongest claims lie in the indirect sex and pregna...
34
[ { "rubric_category": "Structure & Style Rubric", "score_option": 3, "criterion": "Readability - is the text in the memo drafted with line spacings and paragraphs (no more than 3-4 sentences per paragraph) to ensure the overall structure is not too clunky/wordy?", "justification": "This is to ensure ...
There are various key statutes or legal remedies available – but it’s key to actually understand the nuances and differences between the different legal frameworks, and more importantly to incorporate the newly enacted Employment Rights Act 2025 which will expand employment rights in phases during 2026-2027. It will be...
Responses may reference the following sources to ground legal and regulatory claims. Equality Act 2010 Employment Rights Act 1996 Employment Rights Act 2025
This is quite a common situation that lawyers will have to deal when supporting or providing initial legal advice to propsective clients. Furthermore, such a case on flexible working request has also became quite a common work arrangement provided to employees since the covid-19 pandemic.
The task requires synthesizing multipl legal frameworks and timelines before and applying them to Sarah's case. It involves multi-hop reasoning, judgment about the strength of evidence and prospects for Sarah, and the ability to communicate uncertainty clearly to a non-technical audience.
3 - Meets Minimum Standards
There are so many grammatical errors in the prompt, the response and in the rubric. The writing and legal explanations are correct but could be worded much better to provide more clear answers and requests.
4 - Typical
3 - Moderate
There is a need to review the cited materials to help understand the issues and how a court would likely review/proceed.
Well-Specified
4 - Above Standards
The substance is very well done - the legal analysis is completely on point, and the prompt is a relatively narrow legal issue that still requires deep analysis to answer correctly. The gaps are minor grammar issues, which can be easily addressed.
4 - Typical
3 - Moderate
This would likely not be a complex analysis for a UK-based employment litigator, but a non-practitioner based in a different jurisdiction would have a difficult time with this task.
Well-Specified
40
Intellectual Property (IP)
"Persona: You are an in-house attorney that works for a state-licensed cannabis manufacturing compan(...TRUNCATED)
"**_CONFIDENTIAL, PRIVILEGED, ATTORNEY WORK PRODUCT_**\n\n**Memorandum**\nTo: Company P Board of Dir(...TRUNCATED)
43
[{"rubric_category":"Structure & Style Rubric","score_option":3,"criterion":"There at least 3 senten(...TRUNCATED)
"I worked in the cannabis space where this was an often-debated topic for lawyers in the space. Ther(...TRUNCATED)
See citations in Golden Answer.
"5/5: This task closely mirror the type of questions that the board of a company would ask in in-hou(...TRUNCATED)
"3/5: This is moderately difficult because there has not been clear caselaw guidance other than a re(...TRUNCATED)
4 - Above Standards
"The statute citations are not in Bluebook format, and other citations do not include the year or ha(...TRUNCATED)
3 - Somewhat Typical
4 - Hard
"It does not require senior-level expertise--it does require complex reasoning and review of several(...TRUNCATED)
Well-Specified
4 - Above Standards
This is a realistic scenario with competing laws.
5 - Core Activity
4 - Hard
"Requires familiarity with federal, state and patent laws as well as judgement to analyze a murky si(...TRUNCATED)
Well-Specified
82
Regulatory & Compliance
"A mid-sized gambling boat casino in Missouri, which recently launched statewide sports wagering on (...TRUNCATED)
"**Executive Summary**\n\nThe Legal Department for a casino operator (“the Operator”) has been a(...TRUNCATED)
29
[{"rubric_category":"Structure & Style Rubric","score_option":5,"criterion":"Organizes response into(...TRUNCATED)
"As a legal professional working primarily in the casino gambling space, I wanted to come up with a (...TRUNCATED)
"<https://revisor.mo.gov/main/OneChapter.aspx?chapter=313> (Missouri gaming laws)\n\n<https://www.so(...TRUNCATED)
"This is a highly realistic task from the perspective of either an in-house counsel answering a ques(...TRUNCATED)
"I expect this to be a 4 (though closer to the '3' side), because it's exactly the sort of question (...TRUNCATED)
4 - Above Standards
"I loved the fact intensity and the nuanced legal advice. It is limited to one US state (MO) and has(...TRUNCATED)
3 - Somewhat Typical
4 - Hard
"It involved research and analyses of multiple laws and statutory provisions, factual considerations(...TRUNCATED)
Overspecified
5 - Exemplary
"The exercise is very well constructed and enjoyable to read. It needs professional expertise to com(...TRUNCATED)
2 - Uncommon
5 - Very Hard
This exercise requires a specific level of expertise.
Well-Specified

Human Edge — Legal Reasoning Evaluation (Showcase Sample)

A public 6-task sample from a rubric-based legal reasoning evaluation dataset built by Human Edge (humanedgetech.ai). Each task is authored and reviewed by practicing senior lawyers and is designed to produce a verifiable, per-criterion reward signal for post-training and evaluation of frontier language models on high-complexity legal work.

The sample contains one task per legal subdomain, drawn from a larger internal corpus of 50 expert-authored tasks.

Why this dataset exists

Most legal evaluations reduce to multiple-choice recall or bar-exam-style questions. Real legal work does not look like that: it is open-ended, multi-jurisdictional, and judged on reasoning quality, appropriate hedging, and whether the sources actually say what the answer claims they say.

This dataset takes the opposite approach. Every task is a realistic instruction from a practicing lawyer's desk, paired with an expert-authored reference answer and a fine-grained rubric of weighted, binary criteria totalling exactly 100 points per task (29–43 criteria in this sample). The rubric — not the reference answer — is the object of evaluation. That makes scoring reproducible across graders and directly convertible into per-dimension reward signals.

Task architecture

Each row is a triple-component unit:

Component Field What it is
1. Legal reasoning prompt prompt A realistic scenario requiring multi-step reasoning and professional judgment, with role, jurisdiction, and deliverable format specified
2. Golden answer golden_answer An expert-authored reference analysis (17k–38k characters in this sample)
3. Scoring rubric rubrics 29–43 categorized, weighted, binary criteria per task, positive weights totalling 100 points, each with the author's justification

Rubric design

Criteria are binary and weighted. Positive weights reward required content; negative weights penalize specific failure modes (hallucinated authority, overconfident advice, missing caveats).

Every task's rubric is calibrated to the same point budget. Positive criteria total exactly 100 points, allocated across the three positive categories in a fixed split:

Category Sub-category Point budget Criteria What it tests
Substance 65 108
Explicit requirements 27 Did the answer do what was literally asked?
Implicit requirements 27 Did it surface what a competent practitioner would raise unprompted?
Legal correctness 22 Is the law stated accurately?
Reasoning quality 32 Is the analytical path sound, not just the conclusion?
Sources & References 20 34 Are cited authorities real, relevant, and correctly characterized?
Structure & Style 15 32 Does it read like professional work product?
Negative Criteria penalties 25 Penalties for specific, anticipated failure modes
Total 100 199

The point budget holds for all six tasks individually, not just in aggregate — so a model's raw score is already a percentage, and scores are directly comparable across tasks and subdomains despite differing criteria counts. Negative criteria sit outside the 100 points and subtract from the earned total, so a score can fall below zero.

The criteria counts vary by task (29–43) because experts allocated the fixed budget at whatever granularity the material demanded — a task needing many small checks uses more +1 criteria, one turning on a few decisive points uses +5s.

The rubric_category field stores each of these as a single literal string. Filter on these values exactly:

Substance Rubric - Explicit requirements, Substance Rubric - Implicit requirements, Substance Rubric - Legal correctness, Substance Rubric - Reasoning quality, Sources and References Rubric, Structure & Style Rubric, Negative Rubric

Weight distribution across the 199 criteria in this sample: +5 (71), +3 (71), +1 (32), -1 (2), -3 (7), -5 (16).

Contents

Six tasks, one per subdomain:

task_id area Rubric criteria Peer review overall quality (R1 / R2)
18 Employment & Labor 34 3 / 4
40 Intellectual Property (IP) 43 4 / 4
68 Commercial Litigation 34 4 / 5
82 Regulatory & Compliance 29 4 / 5
93 Corporate & M&A 30 4 / 4
100 Contract Law 29 4 / 5

Jurisdictions are US and UK (England and Wales). task_id values are the original corpus identifiers and are intentionally non-contiguous.

Who wrote and reviewed these tasks

Tasks were authored by senior legal practitioners recruited against a hard credential bar: 8+ years in practice, a Master's or PhD in law, and a background in AmLaw 100 / Magic Circle firms, senior courts, government, or in-house at large enterprises.

The contributing cohort averaged ~18.5 years of practice, split roughly 75% US / 25% UK, with education spanning T14 US law schools and leading UK and European universities. Identity was verified for every contributor before any project interaction.

Domain expertise alone does not make a calibrated evaluator, so every contributor completed a mandatory training program (~2.3 hours average) on stress-testing model outputs, applying rubrics consistently, and documenting rationale.

Quality assurance

Three stages, applied to every task in the source corpus:

  1. Automated checks. Every task passes programmatic validation before it reaches a reviewer. Prompts are checked for scenario framing, role specification, deliverable format, and jurisdictional context. Rubrics are checked for binary format compliance and a minimum criterion count. A model-graded pass confirms each rubric is consistent with its golden answer.
  2. Double-blind expert peer review. Two independent senior practitioners review each task, rating overall quality, difficulty, representativeness, and task specification quality, each with written rationale. Reviews drive iterative revision rather than a simple accept/reject: the author revises against reviewer comments, and the version shipped here reflects those revisions.
  3. Difficulty validation. Tasks are stress-tested against frontier models. A task is admitted only if strong models still fail a meaningful share of its rubric criteria — tasks that models solve comfortably carry no training signal and are rejected.

Tasks in the source corpus averaged roughly eleven hours of expert effort each, counting authoring, peer review, and revision. Reviewer ratings clustered tightly, indicating consensus among practitioners rather than averaged-out disagreement.

Field reference

Task content

Field Type Description
task_id int32 Original corpus identifier
area string Legal subdomain
prompt string The legal reasoning prompt given to the model
golden_answer string Expert-authored reference answer
associated_rubrics int32 Number of rubric criteria; always equals len(rubrics)
rubrics list<struct> Nested rubric criteria (see below)
task_details string Author's note on why the task is hard or interesting
reference_materials string Source authorities relied on; free-text, sometimes URLs, sometimes citations

rubrics struct

Field Type Description
rubric_category string One of the seven literal values listed above
score_option int32 Weight: +5, +3, +1, -1, -3, -5
criterion string Binary question applied to the model's answer
justification string Author's rationale for why this criterion matters; empty for 5 of 199 criteria (see Limitations)

Text has been normalized: non-breaking spaces, stray tabs and other exotic whitespace introduced by the authoring tools have been folded to plain spaces. Newlines in golden_answer are preserved.

Author commentary

The task author's own notes on realism and difficulty. Their numeric self-ratings are deliberately not included — an author's rating of their own task is not independent evidence, so the only ratings in this dataset come from the two peer reviewers.

Field Type Description
realistic_explanation string Why the scenario reflects real practice
difficulty_explanation string What makes the task hard, and where models are expected to fail

Peer review

Two independent reviewers per task, both prefixed: peer_review_1_ is the first-round review (conducted on the first draft version), peer_review_2_ the second (conducted on the second draft version). The final version shipped here was produced after revising comments from the second reviewer. Reviewer identities are not published.

The *_label fields carry the rating exactly as the reviewer selected it — score and wording together, e.g. 4 - Above Standards. To get a numeric value, split on the first -.

Field suffix Type Description
overall_quality_label string Rating 1–5, e.g. 4 - Above Standards
overall_quality_description string Reviewer's written rationale for the quality rating
representativeness_label string Rating 1–5 — how typical this task is of real practice, e.g. 5 - Core Activity
difficulty_label string Rating 1–5, e.g. 3 - Moderate
difficulty_description string Reviewer's written rationale for the difficulty rating
task_specification_quality_label string Categorical: Well-Specified, Overspecified, or Underspecified

Rating scales, as presented to reviewers:

  • Overall quality: 1 - Far Below Standards, 2 - Below Standards, 3 - Meets Minimum Standards, 4 - Above Standards, 5 - Exemplary
  • Representativeness: 1 - Rarely Encountered, 2 - Uncommon, 3 - Somewhat Typical, 4 - Typical, 5 - Core Activity
  • Difficulty: 1 - Trivial5 - Very Hard (3 - Moderate, 4 - Hard)
  • Task specification quality: Underspecified, Well-Specified, Overspecified

Personal and sensitive information

The dataset contains no personal data. Scenarios are built on fictional parties and hypothetical facts; the only real names are the litigants in cited case law, which is public record. Author and reviewer identities are not published.

Intended uses

  • Evaluating LLM performance on open-ended, high-complexity legal reasoning
  • Rubric-as-reward research: RLVR, reward model training, process supervision
  • LLM-as-judge calibration — the rubrics give a judge concrete, verifiable criteria instead of a vague quality prompt
  • Studying legal-domain failure modes, especially source and citation reliability
  • A reference template for anyone constructing expert-authored rubric evaluations in other professional domains

Out of scope: legal advice of any kind. The golden answers are evaluation artifacts written against hypothetical facts, not guidance on any real matter, and must not be relied on as such.

Limitations

  • Six tasks. This is a methodology sample, not a benchmark. Do not report aggregate scores over six tasks as a model capability claim.
  • Scoring requires a judge. Criteria are binary but not mechanically checkable; they need a competent grader, human or model. Judge choice will shift absolute scores.
  • US and UK only. Nothing here generalizes to civil law systems, EU-level practice, or other common law jurisdictions without revalidation.
  • Contamination risk. Published openly, these tasks may enter future training corpora. Treat results on this sample as indicative once models trained after its publication are involved.
  • Reviewer identities are withheld. Cohort-level credentials are described above; individual names and affiliations are not published.
  • Five criteria carry no justification. Four in task 40 and one in task 100 have an empty justification; the authors did not record one. The criteria themselves are complete, scored, and counted in the point budget — only the explanatory note is absent. Every other field is populated in every row.

License

Released under the Creative Commons Attribution 4.0 International license (CC BY 4.0).

You are free to share and adapt this dataset for any purpose, including commercially, provided you give appropriate credit to Human Edge, link to the license, and indicate whether changes were made.

The license covers Human Edge's contribution — the prompts, golden answers, rubrics, and review ratings. It does not grant rights in the third-party statutes, regulations, and court opinions cited within the tasks; those remain governed by their own terms. It does not extend to the remainder of the corpus.

Citation

If you use this dataset, please cite it:

BibTeX:

@misc{humanedgeai2026legalreasoning,
  title     = {Human Edge Legal Reasoning Evaluation: Showcase Sample},
  author    = {{Human Edge}},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/HumanEdgeAI/LegalReasoning}
}

APA:

Human Edge. (2026). Human Edge Legal Reasoning Evaluation: Showcase Sample [Data set]. Hugging Face. https://huggingface.co/datasets/HumanEdgeAI/LegalReasoning

About Human Edge

Human Edge builds expert human data for AI development — SME-based evaluation, benchmarking, and reinforcement learning from expert feedback in domains where correctness requires professional judgment: finance, legal, healthcare, and tax.

This dataset is a sample of the pilot phase of a larger program. The production methodology scales the cohort, the review pipeline, and the volume well past what is shown here.

To discuss an evaluation or benchmarking engagement: humanedgetech.ai

Downloads last month
46