Data sources & methodology

Specialty theses are registered publicly. We may keep the thesis number, Turkish and English titles, year, awarding university, specialty, advisor and source-permitted abstract and keywords. A numeric thesis number is not itself a detail URL: Psyduck links only a same-origin official YÖK detail URL that was resolved from an exact-number search and verified against that same thesis number. If no validated URL has been stored, the number remains visible without a fabricated link.

Where this connector is legally approved and enabled, academic title, institution, department, affiliation years, author identifier and public CV publication list may come from YÖK Akademik. Those records help corroborate whether names in separate indexes refer to the same person; a name by itself is never enough for automatic publication attribution.

Approved connectors may combine complementary indexes. Availability and field permissions differ by source:

• PubMed: PMID-linked biomedical records and, where supplied and permitted, abstracts, structured abstract sections, MeSH terms, keywords and affiliations.
• OpenAlex: DOI and OpenAlex work identifiers, citation counts, author identifiers such as OpenAlex Author ID or ORCID, and per-work affiliations from its aggregated scholarly graph.
• TR Dizin: ULAKBİM's index of Turkish journals, including records international indexes may miss; the TR Dizin record number remains visible even where no stable official detail link is available.
• bioRxiv and medRxiv: preprint status and links to a later published DOI. Preprints are explicitly labelled not peer-reviewed, are excluded from the academic topic score, and are hidden when the linked published article is present.

Where this separate evidence source has passed review and is enabled, a condition-specific claim from a practice's official website may be used only with the exact excerpt, official source URL, reviewer and observation dates, verified status and an expiry no more than 180 days after review. Pending, expired, over-age or invalid claims are removed from searchable evidence and cannot establish a direct match. This is sourced professional evidence, not a patient review or an endorsement by Psyduck.

Stable identifiers and permitted professional metadata are kept with provenance. Abstract text, abstract sections, MeSH terms and keywords are stored only when the source permits reuse; otherwise the identifier and permitted metadata remain and the abstract is marked unavailable. A thinner refresh does not erase richer, valid source data already stored.

Linking is conservative because names collide, transliterate inconsistently and appear as initials. Exact identifiers are preferred: PMID, normalized DOI, OpenAlex work or author ID, ORCID, TR Dizin record number and YÖK Akademik ID. Conflicting identifiers stop the write and enter review.

• A name-only candidate never links automatically; medium or ambiguous evidence is held for review and weak evidence is dropped.
• For a thesis-only profile's PubMed cold start, one work is never enough. At least two distinct works must each contain a matching author name and an affiliation compatible with the same active doctor's stored history.
• If either work matches another active doctor, affiliations conflict, or the two-work aggregate is incomplete, nothing is linked. Only the exact PMIDs and DOIs that pass the aggregate gate are attributed, with provenance.

Free text is mapped to a canonical condition and requested facets such as persistence or duration, severity, phenotype, treatment, partial or positive response, and treatment failure. The default result set is strict: a doctor needs direct evidence for the exact condition and the card states which requested facets the evidence does and does not support.

• If strict search has no result, Psyduck returns an honest empty state rather than inserting a generic specialist.
• A broader condition search is run only after the patient chooses that action. A specialty-wide search is a separate, later choice; neither widening happens automatically.
• Direct and related evidence are kept separate, with source labels, snippets where permitted, and links where a validated source URL exists.

Within the requested match tier, results are ordered first by how many requested facets the cited direct evidence supports, then by fused lexical, medical-concept and — only when separately enabled — semantic retrieval relevance. The existing academic topic score is used only as the next tie-break. The direct-search ranker applies no fixed bonus for academic evidence over a current verified official-practice claim, or vice versa; source category alone does not decide the first stages. Numeric scores remain internal and the evidence is shown instead.

Read the full ordering method

A publication count describes confidently linked records, not all work a doctor may have written. Pending, running, needs-review and failed states are not zero; only a completed source check may report no verified record, with its completion date. Coverage can still lag, sources can omit records, affiliations and citation counts age, and conservative matching may leave valid work unlinked. This is a limit of the method, not a judgement about the doctor.

No payment, contract, commercial relationship or advertising budget influences whether a doctor is listed, how they are described or where they appear. There are no sponsored or featured slots. No price or before-and-after image is published. Verified official-practice claims and academic records must meet the same source and relevance rules; neither receives a commercial boost.

If something here is wrong, tell us. Doctors can request a correction or removal at any time, and removal takes effect while the request is being reviewed.

Correction or removal request