An illustration · invented records · the research is real
How we researched this
The search, sources, and review behind the research summary. Use this page to follow a finding back to its source or understand the scope of the work.
Search and preparation
- Run date: September 8, 2026, starting 17:47 UTC. Research dates are recorded separately from the scenario dates.
- Question frozen before searching: question and scope. The exact search strings were shown to Sarah and approved in writing before anything ran.
- Tools: the research-ops evidence pipeline (a set of scripts Sarah maintains), at
code revision
00ed6d4, querying PubMed, OpenAlex, ClinicalTrials.gov, Unpaywall, and Europe PMC. Synthesis and selection were done by an AI assistant (Claude) in a session with Sarah. The run log and check records are kept in the research-ops repository.
What was searched, and what came back
| Step | What | Result |
|---|---|---|
| PubMed | 9 searches, from 2004 onward | 1,102 hits across searches; 730 unique papers; no search was truncated (every hit was retrieved) |
| Query correction | One phrase PubMed silently dropped (“1 to 3 positive”) | Removed before the run completed; the remaining terms in that search were unchanged |
| Foundational-paper check | 5 papers the search had to find or be declared broken | 5 of 5 found by the searches themselves |
| OpenAlex | 5 searches | 3,841 works, used only to find open-access copies and publishing institutions, not for selection |
| ClinicalTrials.gov | 3 searches | 19 adult-eligible registrations; 6 cited |
| Open-access lookup | 3,725 DOIs | 252 of the 373 pruned papers had an open-access link |
| Pruning | Core searches kept whole; side searches capped by recency | 373 papers carried forward |
| Full text | Europe PMC XML, PubMed Central, and the publisher | 130 papers with structured full text; the rest closed or abstract-only |
| Corrections | Errata and retractions | 0 retracted; 23 papers in the corpus carry a published correction, 3 of them cited here |
How the 16 sources were chosen
From the 373 pruned papers and 19 trial records, the assistant read the abstracts and, where available, the full text of the papers that bear directly on the frozen question: the original validation, the two randomized trials in the population, the guideline that interprets them, their long-term and secondary analyses, one independent systematic review, one large real-world study, and the papers that mark the limits (late recurrence, the premenopausal puzzle, core-biopsy testing). The 350-odd papers not cited are mostly single-center series, cost analyses, other genomic tests, and duplicates of the same trials’ findings. The assistant selected the sources; Sarah Kiselycznyk reviewed the resulting notes and summary as described below.
Full papers in this folder
Two papers are included in full as Markdown, under Creative Commons CC BY licenses. Their paragraph locators match the links in the source notes:
- Freeman 2025, the JAMA Network Open study (31 paragraphs, 2 tables).
- Tappenden 2025, the NIHR Health Technology Assessment report (338 paragraphs and 75 tables in the main text, about 31,000 words, plus its nine appendices). Obtained as the publisher’s XML from the NIHR Journals Library on 2026-09-08 and rendered with the same locators.
The remaining source notes provide:
- Eight abstracts. Full text was behind a paywall or could not be retrieved; these source notes use the published abstracts.
- Five full texts read and linked. The two TAILORx papers, RxPONDER, the TAILORx clinical-risk analysis, and the core-biopsy study are free to read as author manuscripts on PubMed Central, and we read them. That copy is free to read, not free to redistribute, so each source note links to it instead.
- One registry note, covering six clinical trial registrations with links to their public records.
Each source note identifies the material used and links to the paper or registry record.
This shareable demonstration includes full papers where redistribution is permitted. A private vault can also organize papers supplied by its owner, subject to their access and usage terms.
Source verification
- Identifiers and retraction status. The pipeline’s citation gate checked all 15 PubMed IDs and 6 trial IDs against the corpus: 0 fabricated, 0 retracted.
- Quotation checks. 37 quotes. For 2 papers the pipeline checked them against the publisher’s XML. For 5 papers the public PMC author manuscript was read and every quote was string-matched against it. For 8 papers only the abstract was available; those quotes were matched against the PubMed record. Result: 37 of 37 matched.
- Quote matching checks the wording against the source. Interpreting a finding also requires its study context; decisions about Mom’s care belong with her and her clinicians.
Review — September 10, 2026
Sarah Kiselycznyk reviewed the research summary and all sixteen source notes for clarity and consistency, with selected quotations checked against their linked source passages. Automated checks matched all 37 quotations to their sources. Source access and correction notices are documented in the relevant notes.
The summary was prepared by an AI assistant using the cited abstracts and full texts where retrieved. The review record retains the scope and changes made during review.
Published corrections
- Three cited papers carry published corrections. Three of the four correction notices were read on 2026-09-10 and none alters a finding cited here: the ASCO 2022 erratum relabels the evidence quality of a recommendation for women 50 or younger; the 2021 Bear 2017 erratum reclassifies one patient without changing the conclusions; the Sgroi 2013 correction narrows confidence-interval lines in figures. Bear 2017’s 2018 erratum could not be opened by our tools and has not been read. Each source note states what its correction changes.
Research scope and source access
- Eight of sixteen sources rest on abstracts alone. Abstracts are the authors’ own summary, but details in the body of a paper can qualify them.
- The search covered English-language indexed literature from 2004. Conference abstracts and unpublished data were not searched.
- This is one bounded question. It does not cover surgery, radiation, endocrine therapy choices, or anything else Mom’s doctors will weigh.
If you want to check a claim yourself
Follow the link beside a finding to its source note. Each quoted passage links to the included paper or a public source record, so you can read the available context. The note also explains whether the full text or abstract was used.