Repository navigation
v1.2.60
- Slide decks get their structure from page images instead of TOC extraction. The coarse profile separates page format from scanning and routes decks to the slide route, which runs once and degrades like TOC extraction.
- Page-memory documents over
page_memory_config.max_pagesare rejected at admission, before billing, PDF conversion and model calls. Documents whose page count cannot be read are rejected there too instead of being charged one page. - Removed the 200-page split (
MAX_PDF_PAGE_LIMITis no longer read), the strict TOC retry, shard residue (ShardPlan, validation and verdict tools,probe.links), and unreachable fragment/JSON parse code.
Reviewed in #566 and verified on staging via #567. PR CI passed, including lint, type checks, tests, CodeQL, and secret scanning. No new migration. The nullable document_page_plan.shard_plan column is kept and no longer written.