Collision Damage Causation and Pre-Existing Damage Distinction QA Dataset
- Records
- 4900
- Size
- 0.3GB
- Format
- JSON
- Provider
MOBIUSI INC- Updated
- v1.02026-11-04
Designed for collision insurance claim assessment, this question-answer dataset combines accident narratives, prior claim and repair records, inspection findings, and claim notes to determine whether disputed damage resulted from the current collision. Each example can include a causation decision, a concise explanation, and an ordered evidence chain, helping models integrate information across sources and explain their reasoning. It supports reasoning SFT and evaluation of damage attribution in vehicle claims.
Overview
Dataset Insights
Preview
Sample Examples
Schema
Technical Specifications
| Field | Type | Description |
|---|---|---|
| file_name | string | File name |
| claim_notes | string | Adjuster or insurer notes relevant to the disputed damage and the claim investigation. |
| evidence_chain | array | Ordered evidence links showing how source observations support or weaken the causation conclusion. |
| accident_narrative | string | Text describing the circumstances, timing, and mechanics of the reported collision. |
| prior_claim_records | string | Text of relevant prior claims, damage reports, and repair history for the vehicle. |
| assessment_conclusion | string | Concise answer explaining the causation determination and any important qualifications. |
| causation_determination | string | Categorical conclusion on whether the disputed damage is attributable to the current collision. |
| disputed_damage_question | string | The question identifying the disputed damage and asking whether it is attributable to the current collision. |
| repair_inspection_descriptions | string | Text describing inspection findings, observed damage, repairs, and any assessments of damage age or condition. |
Trust
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
FAQ
Frequently Asked Questions
- What does the dataset contain?
- Examples focus on disputed damage in vehicle collision claims and combine accident narratives, prior claim records, repair and inspection descriptions, and claim notes. They include a causation question, a determination, an explanation, and an evidence chain.
- Which capabilities can it support?
- The dataset supports training in multi-hop question answering, damage causation assessment, cross-source evidence integration, and conclusion generation. It can also evaluate whether a model can distinguish damage from the current collision from pre-existing damage based on the available records.
- Are the sample count and file format specified?
- The provided information does not specify the sample count or download format. Please refer to the product page or delivery documentation for confirmed specifications.
- How can I obtain the dataset, and what licensing applies?
- The delivery method, permitted uses, and commercial rights are governed by the terms listed on the product page. Confirm the dataset license and any applicable privacy and compliance requirements before use.
Can't find the data you need?
Post a request and let data providers reach out to you.
Get this Dataset
High-quality data, ready for your next AI breakthrough.
Cite this Work
@dataset{Mobiusi2026,
title={Collision Damage Causation and Pre-Existing Damage Distinction QA Dataset},
author={MOBIUSI INC},
year={2026},
url={https://www.mobiusi.com/datasets/ef37939866743d600169621d738cdfe9?dataset_scene_id=3},
urldate={2026-10-21},
keywords={collision damage dataset,vehicle insurance QA,pre-existing damage detection,damage causation reasoning,insurance claim reasoning,accident evidence chain,multi-hop QA data,claim loss assessment},
version={1.0}
}Using this in research? Please cite us.