For years, NCI-funded scientists shared data by attaching spreadsheets to papers. Those files exist, but no catalogue lists them. This tool searches all of them, opens any one in a click, and shows what any NCI grant left behind.
Every square is a paper we opened. 1,076 random NCI-funded papers across four cancers. Lit squares hold a genuine dataset in their attachments: 202 of them. Faint squares hold only small tables.
Three of 9,543.
Verified datasets first.
Europe PMC lists every NCI-funded open-access paper from 2016 to 2022 with supplementary files and no text-mined repository accession: 26,228 papers. We read each paper's attachment list. Nothing downloaded.
Four random samples of 300 papers. Every attachment downloaded and every spreadsheet, CSV, Word and PDF opened. A file counts as a dataset at 50 rows × 3 columns or 1,000 rows.
90 machine decisions re-examined file by file (not blind; verdicts published). All 60 "dataset" calls were real tables. Five other NIH institutes show the same 15–24% pattern.
Every attachment of all 4,759 papers opened: 9,543 datasets, all written into the exact 27-column file the Index of NCI Studies loads. About a fifth (BMC, PLOS) already have a publisher Figshare DOI; we link to those rather than duplicate them.