Research prototype: open-access corpus samples for demonstration; not a live BMJ product or clinical decision-support service.

Live open-content catalogue

Five years of research at two levels of evidence.

Explore real BMJ open-access full text as whole-document evidence for breadth and canonical passage evidence for precise retrieval, citation and provenance.

10,374Open-access articles
40Canonical evidence units
9Completed ingestion jobs
30Named journals
23Discoverable topics

Evidence resolution

Breadth when speed matters. Precision when the passage matters.

Each ingestion generation makes its evidence contract explicit. That lets a client distinguish a one-unit article from a high-resolution article before it retrieves or cites anything.

Passage-level evidence

Example articles — paragraph evidence

Stable, section-aware evidence units are materialised before indexing, giving each passage its own address and locator.

Articles
9
Evidence units
40
Units / article
4.4
Coverage
2025–2026

Generation sample-paragraph-v1

Across ingestion jobs

Six real articles. Six real evidence addresses.

The selection takes the earliest and latest completed job from each populated generation, so it continues to reflect the live corpus rather than a hand-picked fixture.

Passage-level 10 evidence units

BMJ Open Diabetes Research & Care · 5 Sept 2025

TLR5 influences the development of type 1 diabetes

DiabetesEpidemiologyInfectious diseasesWomen’s health
Passage 2 1,851 characters
TLR5 influences the development of type 1 diabetes flagellin.2 In 2011, we found that TLR5 is the only TLR molecule that is upregulated on beta cells following glucose stimulation.3 Initially, the purpose of this upregulation was unclear, but it was hypothesized to be a defense mechanism for beta cells, which are more vulnerable when being active.4 It has been shown that high TLR5 boosts the innate immune system and granulocyte activity.5 TLR5 expression may inhibit T cells, often through induction of regulatory T…
Evidence address
ev_a644e613da61d6ad79cd4398
Ingestion job
job_517989e2dd344b1a9b5a77a64fce22a7
Passage-level 3 evidence units

BMJ Oncology · 2 Jul 2026

Looking inside the primary tumour: what drives immunotherapy response in renal cell carcinoma?

Cancer and oncologyCancer immunotherapyKidney diseaseMedicines and therapeutics
Passage 2 1,892 characters
Looking inside the primary tumour: what drives immunotherapy response in renal cell carcinoma? feature across multiple tumour types and this study adds further support to their relevance in RCC.Equally interesting is what the investigators observed after treatment. Responding tumours showed a reduction in many immune-related transcriptional programmes, whereas non-responding tumours demonstrated increased inflammatory signalling and enrichment of exhaustion-associated pathways. The observation may seem…
Evidence address
ev_666ae6dfea0fc697b78db8b9
Ingestion job
job_885a90d3008e44c1aa623d47c06e50a4

Catalogue shape

What can be filtered before retrieval.

Journal, publication date and controlled topic assignments live in the canonical catalogue, independently of the replaceable search index.

Articles by journal

  • neurintsurg 10,032
  • BMJ Open 308
  • BMJ Medicine 3
  • BMJ Oncology 3

Most represented topics

  • Epidemiology 5 articles
  • Diabetes 4 articles
  • Cancer and oncology 3 articles
  • Diagnostics and biomarkers 3 articles
  • Medicines and therapeutics 3 articles
  • Cancer immunotherapy 2 articles
  • Infectious diseases 2 articles
  • Kidney disease 2 articles
  • Ageing and older people 1 articles
  • Cardiovascular disease 1 articles
  • Child and adolescent health 1 articles
  • Clinical trials 1 articles

Supported query patterns

Questions this proof of concept can ground.

These examples use topics and dates present in the live catalogue. Authenticated clients can combine semantic retrieval with governed corpus scope, filters and stable citations.

01

Evidence discovery

“What does the BMJ open literature report about health inequalities and diabetes?”

hybrid retrievaltopic expansionranked previews

POST /v1/search

02

Change over time

“How did reporting on mental health outcomes change between 2009 and 2026?”

publication windowrolling answercitations

POST /v1/rolling-answers

03

Journal + topic

“Find BMJ Oncology evidence about diagnostics and biomarkers, ordered by relevance.”

journaltopicdate

POST /v1/search

04

Methods intersection

“Find clinical trials that also discuss cancer immunotherapy.”

multiple topicsarticle metadatabounded preview

POST /v1/search

05

Citation-ready passages

“Return passage-level evidence about diabetes, with article, section and stable evidence identifiers.”

paragraph generationsection locatorprovenance

POST /v1/search + GET /v1/chunks/:id

06

Corpus mapping

“Which journals and publication years cover cardiovascular disease?”

cataloguejournal facetsdate coverage

GET /v1/documents

Read the evidence contract

What the two resolutions mean in practice.

Both are useful, but they make different promises. The gateway exposes that distinction rather than allowing it to disappear inside a search result.

Passage-level: citation precision

Each paragraph has a stable evidence ID, section locator, content hash and bounded source object. A result can point to the exact unit used for retrieval or answer grounding.

Document-level: corpus breadth

One evidence ID represents the full article. Managed search chunking can still find relevant text, but the canonical gateway citation resolves to the article-level unit rather than a stable paragraph.

Move from the showcase to the governed API.