# Reproduce the Internet Archive download checks Checked September 11-12, 2026. This is original inspection methodology for the [Internet Archive download guide](/blog/download-internet-archive-books), not a distribution of the books. Obtain source files from the provider's normal public download options, subject to their current availability and terms. ## Source records - [Candide item: 1761 catalog record](https://archive.org/details/candideouloptimi00volt_1) - [1761 item metadata and exact file list](https://archive.org/metadata/candideouloptimi00volt_1) - [Candide item: 1759 catalog record](https://archive.org/details/candideouloptim01voltgoog) - [1759 item metadata and exact file list](https://archive.org/metadata/candideouloptim01voltgoog) The metadata was refreshed on September 12. The EPUB and DjVu were downloaded on September 11; their size, SHA-1, and MD5 match the refreshed records. The PDF and scandata were downloaded and hash-checked on September 12. The public [inspection record](inspection-results.json) records SHA-256 hashes to identify the inspected bytes. A different hash means you are inspecting a different file. ## EPUB package check Save the ordinary offered `candideouloptimi00volt_1.epub` and the [read-only inspector](inspect-epub.py) in a working folder. Python's standard library is enough; the recorded run used Python 3.14.3. ```bash python3 inspect-epub.py candideouloptimi00volt_1.epub ``` Expected observations for the recorded hash: 344 ZIP entries, a null bad-CRC entry, 341 manifest items, 280 spine entries, empty package title and language lists, and zero links in `EPUB/nav.xhtml`. These are separate observations, not an overall EPUB pass/fail score. The inspector does not run EPUBCheck, render an ebook, follow every link, proofread text, or certify completeness. To read one of the original text samples without modifying or unpacking the archive to disk, run: ```bash unzip -p candideouloptimi00volt_1.epub EPUB/page_260.html ``` Use `EPUB/page_11.html` and `EPUB/page_100.html` for the other body samples. The inspected HTML files declare their language as `None`; this is not a verified language value. The catalog identifies the book as French. ## Same-edition PDF and visible-text comparison Save the item's ordinary `candideouloptimi00volt_1.pdf`. With Poppler installed, run the following; the recorded run used Poppler 26.08.0 for all three commands: ```bash pdfinfo candideouloptimi00volt_1.pdf pdftoppm -f 260 -l 260 -scale-to 1600 -singlefile -png candideouloptimi00volt_1.pdf sample-260 pdftotext -f 260 -l 260 -layout candideouloptimi00volt_1.pdf - ``` Open `sample-260.png` and compare it with the extracted text and EPUB HTML. The visible printed page is 252; the lines containing Turquie and Pangloss are readable in the image but corrupted in both extracted versions. This comparison is visual; do not infer image accuracy from the extracted text. Repeat at PDF file pages 11 and 100. Page 11 shows an ornamental chapter opener; spurious characters appear before the title in the text versions. File page 100 is visibly numbered 92. We also rendered file page 9 to inspect the title page and its date, M. DCC. LXI. The public `candideouloptimi00volt_1_scandata.xml` file has 278 scan records, 276 marked `addToAccessFormats=true`. That included count agrees with this PDF's 276 pages. Our sampled PDF file pages map to scan leaves with the same numbers. This correspondence is item-specific; do not apply a universal offset to other books, or treat matching counts as proof that the printed volume is complete. ## Second record: availability only The September 12 file list for `candideouloptim01voltgoog` had a DjVu entry and no `.pdf` or `.epub` entry. We downloaded the DjVu and checked its hashes, but did not decode its pages, evaluate its OCR, or convert it. Its catalog date alone does not prove it is the appropriate edition for a reader's task. ## Scope and reuse No whole third-party books or page images are included in these public assets. The CSV worksheet, inspector, method, and observation summaries are original BookTranslator materials. No customer material was used. No BookTranslator translation, device benchmark, whole-book error rate, or rights determination was performed. Availability and regenerated files can change after this check.