Repository Issue Activity (beta)

#1564 docs: Documentation website makes it unusable

#1568 feat: Enable page-level confidence scoring

#1567 chore(lint): pay down the size and complexity baseline

#1565 bug: PDF: one single-column line vetoes the whole row split, so a ruled troubleshooting page becomes one table and its section headings become cell text

#1566 bug: PDF: a word cut across two touching spans is rejoined with a space, so `prijs` is indexed as `pri js`

#1562 bug: DOCX: a text box drops every XML entity reference, so a cover-page title silently loses its `&`

#1561 bug: PDF: a suspended hyphen mid-line is dehyphenated, welding a noun to the conjunction after it

#1560 bug: PDF: a 1.08 pt baseline offset breaks a table row apart, and the fragments come back reordered and joined without a separator

#1558 bug: PDF: the header/data split of a reconstructed table DELETES rows, so a parts list loses its first priced articles

#1559 bug: DOCX: a page whose inline images cannot fit gets no break, so every page after it is one too low

#1557 bug: Multi-line code samples in PDFs lose all line structure in markdown output (no fenced code blocks)

#1556 bug: Character/glyph substitution corrupts identifiers in PDF text extraction

#1555 bug: PDF: a drawn row band is subdivided by the baselines of the wrapped cell inside it, inflating the grid until it is rejected as prose

#1554 fix(ocr): load_image_for_ocr decodes under the default security limits, not the configured ones

#1256 Your project is on StackMap — a curated map of the AI stack

#1541 bug: Installing xberg via PIE gives error on Debian 13

#1553 bug: legacy `.doc` headings are detected by line length rather than read from the document's styles

#1552 bug: legacy `.doc` elements are split on blank lines, merging consecutive Word paragraphs into one element

#1550 bug: legacy `.doc`: Word's automatic list numbering is dropped entirely — an auto-numbered clause arrives as bare prose, while the DOCX path keeps it as a `ListItem`

#1551 bug: legacy `.doc`: `fcClx` is read at `FibRgFcLcb97` index 66 instead of 33, so the piece table is never used and the fallback decodes CP1252 as UTF-16LE

#1548 fix(native-pdf): the Latin-1 fallback warns per span for a condition it handled correctly

#1549 bug: DOCX: a cell merged across N columns comes back N times — `gridSpan` is expanded by cloning the cell, `vMerge` by copying the row above

#1547 bug(pdf): recoverable input handling is logged per occurrence at WARN

#1545 bug: PDF: a multi-column table beside a prose column is assembled in full-width Y order, welding the table's own columns together

#1543 bug: PDF: a real hyphen that ends a line is deleted when the lines are joined