6ljxSrSSKJr SSKJr SSKrSSKJr SSKJ r Sr S Sjr S S S jjr \ S :Xa\ "5 gg)uBackfill `financebench.documents.page_count` from the PDFs themselves. The FinanceBench release files carry no page counts, so this reads each document's PDF from the bucket (through the quber S3 cache) and records its page count. Run it after `load.py` — the loader rebuilds the schema, which clears the column. A missing PDF logs a warning and leaves the count NULL; it is never a silent skip. ) annotations)OptionalN)logger)dbz qubera-docsc 6SSKJn [R"5nUR S5R 5nS=p4UH\upVU"S[ SU35n[R"U5nURn SSS5 UR S W U45 US- nM^ SSS5 [R"S WW5 X44$!,(df  NN=f![a(n [R"SXVU 5 US- nSn A MSn A ff=f!,(df  No=f) zbSet page_count for every document whose PDF is in the bucket. Returns (updated, missing) counts. r)resolve_documentzESELECT doc_name, s3_key FROM financebench.documents ORDER BY doc_namezs3:///NzNo page count for {} ({}): {}zEUPDATE financebench.documents SET page_count = %s WHERE doc_name = %sz2Page counts: {} documents updated, {} missing PDFs)quber.files.cacherrconnectexecutefetchallBUCKETpymupdfopen page_count Exceptionrwarningsuccess) rconnrowsupdatedmissingdoc_names3_keylocalpdfnexcs >/home/mande/repo/quber/src/quber/playground/benchmark/pages.pybackfillr!s 3 || S (*  $ H (5&)BC\\%(CA) LLWH   qLG!% ( NNGRYZ  )( >RUV1   sY*D (C/ C<CD  C CD C DD<D DD  Dc[5 gN)r!)argvs r mainr%5s J__main__)returnztuple[int, int]r#)r$zOptional[list[str]]r(None)__doc__ __future__rtypingrrlogururquber.playgroundrrr!r%__name__r&r r1s># > zFr&