Contradictory Case File Evidence Retrieval Benchmark

#counterevidence retrieval #contradiction detection #evidence-based reasoning #legal case evidence retrieval #conflicting testimony analysis #legal document review

5000
0.5GB
JSON
CATL
MOBIUSI INC

Updated:2026-10-18

AI Analysis & Value Prop

Designed for fact verification in commercial litigation, this benchmark pairs query claims with legal document excerpts that support, contradict, or only superficially relate to the issue in dispute. Annotations identify evidence relationships and counterevidence types, with source metadata, source anchors, and rationale. The data supports evaluation of whether models can find meaningful evidence among similar or conflicting narratives, assess whether a claim is undermined, and trace conclusions to the relevant case file.

Dataset Insights

Sample Examples

{}

Technical Specifications

Field	Type	Description
file_name	string	File name
query_claim	string	A commercial litigation fact or legal claim to be retrieved and verified.
source_anchor	string	A concise citation, key sentence, or verifiable location note that anchors the judgment to its case file source.
relation_label	string	The primary evidence relationship between the candidate excerpt and the query claim.
source_metadata	object	Case, document, and original-text location information associated with the candidate excerpt.
is_claim_refuted	boolean	Indicates whether the candidate excerpt directly or materially weakens or refutes the query claim.
candidate_excerpt	string	A candidate evidence passage retrieved from a case file document for assessing the query claim.
annotation_rationale	string	Explains the semantic relationship between the excerpt and query claim, including the point of conflict or lack of sufficient support.
counterevidence_type	string	The specific type of counterevidence represented by the excerpt, such as refutation, conflicting account, or insufficient support.
annotation_confidence	number	A score from 0 to 1 representing overall confidence in the evidence relation, counterevidence type, and source anchoring judgments.
retrieval_relevance_score	number	A score from 0 to 1 indicating the candidate excerpt's evidentiary relevance to the query claim.

Compliance Statement

Authorization Type	Proprietary - Commercial AI Training License (No Redistribution)
Commercial Use	Requires exclusive subscription or authorization contract (monthly or per-invocation charging)
Privacy and Anonymization	No PII, no real company names, simulated scenarios follow industry standards
Compliance System	Compliant with China's Data Security Law / EU GDPR / supports enterprise data access logs

Frequently Asked Questions

What does the benchmark contain?: Each sample links a commercial litigation query claim to candidate legal document excerpts and provides labels for evidence relations, counterevidence type, whether the claim is refuted, source anchors, and annotation rationale. Check the product page for dataset size and file specifications.

Which capabilities can it evaluate?: It can evaluate retrieval from similar or conflicting legal text, identification of counterevidence and conflicting accounts, recognition of relevant but insufficient excerpts, and attribution of conclusions to verifiable case file sources.

Can the data be used to train models?: The annotations can support training or evaluation for evidence retrieval, evidence relation classification, counterevidence identification, and source attribution. Review the data format, intended scope, and annotation conventions against your task before use.

How can I obtain the data and check its license?: Consult the product page or the data provider's published channels for access methods, available formats, and license terms. Commercial use, redistribution, and any restrictions related to case file materials are subject to the applicable license and data compliance requirements.

Can't find the data you need?

Post a request and let data providers reach out to you.

Get this Dataset

Verified for Enterprise Use

Cite this Work

@dataset{Mobiusi2026,
  title={Contradictory Case File Evidence Retrieval Benchmark},
  author={MOBIUSI INC},
  year={2026},
  url={https://www.mobiusi.com/datasets/dc044a8c5085b2a5fdf61b22bbbf82a0?dataset_task_cate_id=10},
  urldate={2026-10-15},
  keywords={commercial litigation dataset,legal evidence retrieval,counterevidence identification,case file document analysis,legal text retrieval benchmark,evidence relation classification,claim fact verification,legal AI evaluation},
  version={1.0}
}

Using this in research? Please cite us.

placeholder

Products

Scene

Domain

Modality

Task