ASKIP (African Sovereign Knowledge Infrastructure Program )

ASKIP — African Health Evidence Corpus (Gold tier, facts-only) v1.1.0  structured evidence items extracted from African biomedical and institutional literature — each one a single quantified claim, with its numeric value, the diseases and countries it concerns, its temporal context, and its full provenance chain back to the source publication. Built to make African health knowledge queryable rather than merely readable. Coverage 42 African countries — Nigeria (13 119), Kenya (10 854), Gabon (4 880), Cameroon (2 607), Mali, Burkina Faso, Senegal, South Africa and 34 others Bilingual — 47 348 English, 15 877 French. The francophone segment is the project's differentiator: rarely available at this scale elsewhere Major disease areas — malaria (9 001), HIV/AIDS (5 535), sickle cell disease (2 618), diabetes, COVID-19, tuberculosis, hepatitis, and hundreds more Sources — PubMed, Europe PMC, OpenAlex, HAL, WHO Africa, Africa CDC, national health ministries, DHS surveys What's in the deposit askip_evidences_gold_v1.jsonl / .csv — the master file, 63 227 rows rdf.zip — 865 449 Turtle triples with PROV-O provenance, ready for SPARQL qa.zip — 3 372 question-answer pairs for LLM evaluation and fine-tuning croissant.json and schemaorg.jsonld — machine-readable metadata for dataset discovery 14 corpus descriptions and 11 thematic syntheses by country and disease DATA_CARD.md and METHODOLOGY_REPORT.md — full documentation Use cases Evidence retrieval — find every documented prevalence for a given disease and country, with its source Knowledge graph work — the RDF export ships with entity normalisation and a curated concept referential LLM training and evaluation — QA pairs and structured claims grounded in traceable African sources Health-systems research — compare what is documented across countries, and where documentation is absent Method and transparency Extraction is automated (LLM), validation is mechanical: structure, plausibility, deduplication. No line has been reviewed by a human. The data card documents every known limitation with an explicit status — counted, estimated, read, or not measured — including the three that matter most and the seven questions we have not answered. Research use. Verify against primary sources before any clinical or public-health decision. Portal: askip.e-shepha.com