Doc2X PDF size splitting and large-page compression — 2026-10-13

Input: pdf-split-size-input.pdf (1,941,743 bytes, 8 pages), authored locally.
Seven small text pages and one large simulated scan (file page 3) deliberately
create uneven page sizes. Texture increases image bytes; this is a reproducible
stress example, not a customer's scan or a typical compression benchmark.

Actual current public product operations:
1. PDF split: size target 1 MB, tolerance 10, naming size-1mb-{000}.pdf.
   Native ZIP contains pages 1–4 (1,936,730 bytes) and 5–8 (3,521 bytes).
   The first result exceeds both 1,000,000 and 1,048,576 bytes.
2. Same input, Equal / by number of parts / 3; clicked split again.
   Native ZIP has page groups 1–3 (1,936,020 bytes), 4–6 (2,850 bytes),
   7–8 (2,100 bytes). Equal page counts do not mean equal file sizes.
   The earlier filename pattern persists; names do NOT encode this operation.
3. Same input, page range 3, clicked split again, downloaded the single file.
   Native extracted page: pdf-split-large-page.pdf, 1,934,664 bytes.
4. Selected that unchanged native extracted page in PDF compression.
   Standard (75% quality), all pages; all three default structure/title switches
   remained on. Native output: pdf-split-large-page-compressed.pdf, 772,682 bytes.

Every source page in the two split ZIPs and the native single-page extraction
matches the original text, decoded page content stream, geometry, rotation and
extracted image bytes. No page is omitted or duplicated in either full split.
No bookmark, form, password, signature, scan recognition or OCR behavior was
tested. All four output download events were observed, one click per download.

Compression retained one 595 x 842 point page, and the title/footer text remains
equal after trimming only final whitespace (the original had one final newline).
The 1400 x 1800 image is re-encoded from Flate to JPEG; this is lossy compression.
Numbers 18, 28, 20, total 66, chart bars and labels were visually checked after
rendering the actual files. No universal quality or size guarantee is implied.
The chart text was already inside a raster image and was not made selectable.

The UI displayed tiny valid split files as 0.00 MB. Exact nonzero sizes and real
page contents are recorded in the checks; the ZIP's size is not each PDF's size.

This proof ZIP was assembled locally; the two inner split ZIPs and PDFs are
unchanged native downloads. Page previews are separately labeled local renders.
Local single-page controls used during input authoring are NOT product output
and are excluded from this bundle.

中文：8 页自编样例中仅第 3 页含大幅模拟扫描图。按大小、均分、提取单页和
标准压缩均由当前工具实际执行。两份原始切分 ZIP、单页 PDF 与压缩 PDF
保留原始下载字节。按大小不能保证每份低于目标；均分只按页面组织。
标准压缩将单页降至 772,682 字节，但图像为 JPEG 有损重编码，需核对自己的资料。
