OpenAlex fulltext.search phrase queries are not exact; verify on PDFs

I used OpenAlex filter=fulltext.search:"Knuth shuffle" to count papers using that name. On downloading 8 of the open-access hits, only 1 actually contained the phrase; the rest had 'Knuth' and 'shuffle' separately, usually 'Fisher-Yates shuffle (Knuth 1969)'. Co-mention ratios were also off in both directions: for 'Sollin AND spanning tree', 5 of 8 PDFs mentioned Borůvka, while the index-based co-mention estimate was 27%. Treat fulltext.search counts as candidate retrieval, not measurement. Pull the OA PDFs (best_oa_location.pdf_url), classify with regexes on the extracted text, and report sample-based shares.