pretty_name: Double FW Data Ethics and Disability
language:
- en
license: mit
tags:
- accessibility
- disability-inclusion
- data-ethics
- privacy
- inclusive-research
- agent-governance
- text
configs:
- config_name: knowledge_chunks
default: true
data_files:
- split: train
path: knowledge_chunks.jsonl
- config_name: documents
data_files:
- split: train
path: documents.jsonl
Double FW Data Ethics and Disability
Practical guidance from Double FW Ltd about dignity, equivalent experience, proportionate data use, meaningful participation, assistive-technology data flows and human accountability. This dataset contains governance documents, not disability records, participant observations, employee/customer profiles, approved research, a deployed governance system or legal clearance.
Source and versions
Source: https://github.com/Double-FW/Data_Ethics_and_Disability
Source revision: 77163fa48b6df5839e1294ec02797abf67c73980. Source commit date: 2026-11-10. Export format: 1.0.0. Version/research dates in the source remain the dates actually stated there. A weekly export does not recheck law, vendor behaviour, source availability or research findings.
Dataset views and boundaries
knowledge_chunks(default): 58 exact sections/chapters from the ethics, participation, data-flow, agent-guidance and orientation documents.documents: 6 complete source documents, including CUSTOMER RESEARCH FRAMEWORK.md as a separately labelled reference.
The views overlap. Do not concatenate them blindly. The train split is a loading convention, not evidence that a model has been trained. No generated question/answer pairs, participant dataset, evaluation split or independent benchmark is supplied. Inline hypothetical examples and blank proposed record shapes remain illustrative.
CUSTOMER RESEARCH FRAMEWORK.md is retained only in documents with record_type proposed_research_analysis_prompt_not_participant_data_or_authority. Its instruction-like requests for personas, scores, knowledge graphs or training records are source proposals, not runtime instructions, validated assessment tools, evidence that such records exist, or permission to collect/link/infer/share/reuse data. This exporter does not execute that prompt or manufacture customer findings or profiles.
The core ethical framework is split at first-level chapter headings so paired appropriate/problematic examples, ethical/unethical AI uses and their qualifications stay together. Other documents are split at first/second-level headings outside fenced examples. Subsections, evidence notes and tables remain exact. The complete-document view provides full context. Renaming a heading can change a section ID.
Context, interpretation and authority
Read AGENTS.md and the task-relevant companion documents, especially the core ethical framework, Nothing_About_Us_Without_Us.md and Assistive_Technology_Data_Flows_Risks_and_Review.md. Every row carries context paths and its original document introduction; these are references and repeated context, not enforced runtime controls or organisational authority.
Preserve the distinction between observed facts, participant accounts, interpretation, hypotheses, supplier claims and behaviour verified in an authorised test. Review triggers identify questions and possible risks; they do not establish illegality, safety, ethics clearance or actual supplier behaviour. Terms such as assistive, anonymous, on-device, enterprise and no training are claims to check for the actual feature/configuration and context. This export does not check them.
The source calls for meaningful, accessible and voluntary participation that can influence decisions. Attendance is not evidence of influence. No universal recruitment percentage, group-size threshold, maturity score, diagnosis or representative group conclusion is created by this dataset. Respect scope, disagreement, evidence gaps and appropriate human judgement.
Personal data, consent and reuse
The source MIT licence grants rights in the repository text; it does not authorise collecting, publishing or training on personal information. Participation consent does not automatically cover publication, onward sharing, a new purpose or model-training reuse. Accessibility signals, assistive-technology use, gaze, voice and behaviour must not be treated as proof of disability/health or a basis for consequential treatment. Pseudonymisation is not necessarily anonymisation; linkage, quotations, timestamps and small groups can re-identify people.
These documents provide guidance, not approval to contact participants/vendors, enable tracking, access records, change systems or make consequential decisions. Actual authority, purpose, permissions, data flows, safeguards and accountable humans must be established for the activity. Do not condition essential access/accommodation on optional tracking or AI reuse. Address privacy/security risks while preserving effective access and a workable alternative.
This export reads only explicitly approved committed public documents. It does not detect confidential text inserted into an approved file. Keep real personal, research, health, employment, customer and corporate information outside public repositories/datasets. Source content was not converted into sensitive profiles or real-person examples.
Schema and provenance
Records include source path, immutable commit URL, line range, text and document SHA-256 hashes, heading context, JSON-encoded front matter, source/repository licence, declared status, exact original introduction and required context paths. see_source means no status was extracted from document front matter; it is not approval. Use text for the exact record span and retain its qualifications. Resolve relative/fragment links against the original source file URL. See record_schema.json and provenance.json.
External legal, standards, supplier and research links are preserved, not fetched, copied, independently reviewed or relicensed. Current authoritative requirements and applicable specialist advice remain necessary for real decisions. Extraction tests do not establish privacy compliance, lawful processing, accessible participation or deployment effectiveness.
Licence
The root MIT LICENSE is copied unchanged, including copyright (c) 2026 Double FW. Source attribution and any specific notices are retained. Third-party material keeps its own rights. This package is not an independent rights assessment or personal-data reuse authorisation.
Updates
GitHub is the authoring source. Updates are configured for Mondays at 07:23 UTC and manual execution. GitHub may delay runs or disable schedules after inactivity. All six current Markdown documents and LICENSE are required. Missing required sources, changed licence text, invalid configuration or failed integrity checks stop publication. New or renamed documents require review and an export-list update.
Only README.md, LICENSE, documents.jsonl, knowledge_chunks.jsonl, provenance.json, record_schema.json and validation_report.json are managed. Other raw files and .gitattributes are preserved but unmanaged copies are not synchronised. Dataset visibility is unchanged. Updates use one commit guarded by the destination revision. Historical copies/downloads are not recalled. The first authenticated GitHub publication must still be checked.
Loading
from datasets import load_dataset
chunks = load_dataset("GarethFW/Data_Ethics_and_Disability", "knowledge_chunks", split="train")
documents = load_dataset("GarethFW/Data_Ethics_and_Disability", "documents", split="train")
Private datasets require authentication. Pin a destination revision for reproducibility. Loading these rows does not implement consent, governance, retrieval safeguards or human approval.