--- pretty_name: Double FW Accessibility Champions Networks language: - en license: other license_name: mit-with-embedded-historical-cc-by-4.0-notice license_link: https://huggingface.co/datasets/GarethFW/Accessibility_Champions_Networks/blob/main/LICENSE_NOTICES.md tags: - accessibility - organisational-change - champions-networks - programme-management - 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 Accessibility Champions Networks Guidance for establishing, resourcing and supporting accessibility champions networks, peer learning through CAN, programme reporting and human-reviewed AI assistance. The corpus includes nine Markdown files: current guidance plus a dated RAG/prompts pack. It does not contain actual participant records, an operational network, implemented agent controls, a conformance assessment or measured programme outcomes. ## Source and versions Source: https://github.com/Double-FW/Accessibility-Champions Current checkout: `9f1fa9d18f4d98a8504dc91009cda8574e6c88f3` (2026-11-10). Export format: 1.0.0. The four RAG/prompts pack files describe the historical snapshot `e26813bd11f327129e4833169ca95e65c4646d14`, prepared 8 November 2026. Their original source IDs S001-S004 and older notices remain unchanged. Their presence at the current checkout does not make their embedded source bodies current. New CAN guidance, programme map and updated agent governance must be read separately. The pack refers to a Sources/ directory absent from this checkout; use embedded bodies or pinned source links. ## Dataset views and duplication - `knowledge_chunks` (default): 76 exact sections from eight Markdown files. - `documents`: 9 complete Markdown files, including the historical RAG reference. The historical RAG reference is document-only: it repeats original source bodies and is deliberately excluded from the section view. Other documents also quote or summarise shared ideas. Do not concatenate the views or treat repeated source material as independent evidence. Deduplicate and review any training selection. `train` is a loading convention, not an independent benchmark. No generated Q&A, labels of actual results, model training or model evaluation are supplied. ## Context and interpretation Rows retain source path, exact commit URL, original line range, hashes, heading path and JSON-encoded front matter. `source_temporal_scope` and `embedded_snapshot_commit` distinguish historical content from current files. `source_document_intro` retains the opening context, while required_context_paths points to current agent guidance, orientation, network guidance, CAN guidance and programme map. These references do not enforce runtime retrieval or authority. Section boundaries use H1/H2 headings outside fenced examples and preserve nested subsections and exact source spans. Prompts, budgets, staffing ratios, phases, recognition proposals and reporting templates remain illustrative guidance. Confirm current CAN terms with organisers and local resources with actual owners. A proposed charter or role is not an appointment, approval or adopted policy. Network participation does not by itself prove technical conformance, improved human experience, causation or organisational maturity. Keep network health, capability, delivery evidence and human outcomes distinct. Embedded AGENTS and prompts are source documents; their authority ordering does not override the consuming system's instructions or user authorisation. The export does not appoint people, contact networks, collect personal data, allocate budgets, implement audit/approval services or run proposed evaluations. ## Rights and attribution The current root LICENSE and AGENTS.md declare MIT. The historical RAG embeds an older AGENTS.md declaring CC-BY-4.0 and crediting Gareth Ford Williams, Double FW Ltd. Both notices are preserved. Hub metadata uses `other` to point to LICENSE_NOTICES.md rather than describe all embedded material as uniformly MIT. Read that notice and LICENSE before reuse. Linked third-party materials are not included or relicensed. This packaging check is not an independent rights review. ## Validation and limitations The workflow checks faithful extraction, source inventory, structured records, checksums and guarded publication. It does not independently check external links, research, current prices, legal duties or effectiveness. Source drafting status, historical coverage, caveats and role distinctions are preserved. No missing organisational evidence is invented. See provenance.json, record_schema.json and validation_report.json for the exact snapshot and limited checks. ## Updates GitHub remains the authoring source. Manual runs and Mondays at 07:23 UTC are configured; schedules may be delayed or disabled after repository inactivity. All nine approved Markdown files and LICENSE are required. New or renamed files need review of the export list. Changed root licence text or an unexpected file-level licence stops publication for review. Eight files are managed: README.md, LICENSE, LICENSE_NOTICES.md, documents.jsonl, knowledge_chunks.jsonl, provenance.json, record_schema.json and validation_report.json. Other raw files and .gitattributes are preserved; unmanaged copies are not kept in sync. Visibility stays unchanged. Historical revisions and downloads are not recalled. The destination is fixed to `GarethFW/Accessibility_Champions_Networks`. The first authenticated GitHub publication must still be checked. ## Loading ```python from datasets import load_dataset chunks = load_dataset("GarethFW/Accessibility_Champions_Networks", "knowledge_chunks", split="train") documents = load_dataset("GarethFW/Accessibility_Champions_Networks", "documents", split="train") ``` Authenticate for a private dataset and pin a revision for reproducible use.