Convert DOCX, PPTX, XLSX/XLS/CSV, ODT/ODS/ODP, EPUB and HTML files (URLs or key-value store) into clean Markdown with headings and tables kept, plus RAG chunks, sheet JSON, speaker notes and metadata. Zip-bomb safe, no LLM needed.
Billing/UX: when the run budget (maximum total charge) no longer covers a document, that document and every document still queued now get a skipped_budget row (charged: false) instead of disappearing silently. The status message counts them.
A large document (large-document-converted) that does not fit the remaining budget no longer stops the run: it is skipped with a skipped_budget row and smaller documents that still fit are converted.
Tables (DATA-1): merged cells no longer shift data under the wrong column. Vertically merged cells (DOCX w:vMerge, HTML/EPUB rowspan) and rows that start part-way across the grid (DOCX w:gridBefore) were dropped, so the cells after them moved one column left. Now a merged value is written once, in its first (top-left) cell, and every position it covers is an empty cell, so each column stays under its own header (same convention already used for horizontal merges, PPTX, XLSX and ODF). DOCX results with merged cells or nested tables get an explicit note in warnings.
