PDF

The text stream in document formats such as PDF is not always stored in visual order; multi-column layouts, mixed text and image arrangements, and interleaved footnotes and headers can cause the underlying text stream to be misaligned with visual order.

The PDF format is a page description language released by Adobe in 1993, designed with the goal of accurately reproducing the visual appearance of documents rather than preserving semantic structure.

PDF redaction that merely covers text with a black rectangle does not truly delete the underlying content, which can be restored by copying the text or removing the layer.

PDF is fundamentally a print-oriented format, focused on where pixels appear on a page rather than the logical structure of text.

PDF内部结构是基于PostScript的页面描述语言，包含字体嵌入、图层、加密等复杂元数据，需要通过PyMuPDF、pdfplumber等专用解析库才能提取可读文本

The underlying format of PDF files does not store logical reading order; it only records the absolute coordinate position of each character on the page.

PDF文档可以被建模为强化学习环境：状态是已读页面与查询的组合，动作是翻页或检索，转移由文档结构决定，奖励来自答案正确性。

A PDF is not text with coordinates but a visual argument about how ideas relate; flattening it into Markdown is reconstructing the author's intent.

双栏PDF中多数提取器按行从上到下工作，导致左栏第一行后紧跟右栏第一行，分块混入两段无关内容。

In addition to visible content, PDF files may also contain text layers, annotations, form data, bookmarks, and even hidden historical revision information, so handling only the visible portion is often not enough.

PDF is designed to keep document layout rendering completely consistent across different devices, operating systems and software.

PDF can contain multi-page content, embed fonts, preserve vector graphics, and support text search and copying.

In office and administrative scenarios, most online service systems, reimbursement platforms and contract signing tools require or preferentially support submission in PDF format.

Out-of-the-box generic text splitters blindly cut off PDF content from the middle of sentences, or slice multi-row financial tables down the middle, destroying header context.

An open password corresponds to the "encryption dictionary" and "user password" in the PDF specification; the file content is genuinely encrypted, and without the password only garbled text can be seen.

Permission restrictions correspond to the "owner password" and permission flag bits in the PDF specification; the file content itself is not encrypted, and any tool capable of parsing the PDF structure can bypass the restrictions and directly extract text or images.

You can view a PDF file's encryption method and key length in Adobe Acrobat under 'Document Properties—Security.'

If only a permission password is set without an open password, the PDF file content is not encrypted, and online decryption services or script tools can easily bypass print and copy restrictions.

40-bit RC4 encryption has been proven to be brute-forceable within hours.

The PDF was generated by a scanner-style pipeline: pages were first rasterized into images, then an invisible OCR text layer was overlaid for accessibility, and the extractor read the vector outline layer (titles, bookmarks), so it returned only the title.