The Last Mile of Public Dataset Discovery: Polly Scout vs. Claude's Native Search
Key Highlights
Elucidata benchmarked two approaches to finding public biological datasets, Claude's native PubMed and web search against Polly Scout, a custom MCP and skill, across two cases, Oral Lichen Planus and Kabuki syndrome, scored against a 16-entry reference list.
Claude's default search sees only PubMed abstracts and titles, with no access to GEO sample-level metadata, so recall fell as the cases got harder.
Polly Scout connects Claude to Elucidata's indexed corpus of 260K GEO datasets and 7.1M full-text PubMed articles, adding automatic synonym expansion, multi-source enforcement, and a completeness-critic pass.
As a result, Polly Scout delivered 2.5 to 3.3x higher recall than Claude across both cases, climbing to 87.5% recall once low-confidence candidates are counted, versus 19-25% for Claude, while holding 100% precision on every presented result.
.svg)
.svg)
%201%20(1).svg)
.svg)
%201.svg){kind=icon}
.svg)
%20(1).svg)
