Built so you don't have to trust us.

Every artifact — the raw response, the extracted text, the rendered DOM, the screenshot, the PDF — is stored under the SHA-256 of its own bytes. The hash is the filename, so altering a file changes where it lives. Nothing can be edited in place without announcing itself. Each hash is printed beside the version it belongs to, and every download bundle ships with a SHA256SUMS you can check offline.

Our copy shows what we saw. A copy at the Internet Archive shows a neutral third party saw the same thing. Every captured URL is submitted to the Wayback Machine and their timestamp is filed next to ours. If this site vanishes, the evidence does not.

Where a company publishes a PDF, that file is the binding artifact and it is offered exactly as they served it, marked as theirs. Where we print a web page to PDF ourselves for convenience, that is marked as ours. The two are never presented as the same thing, because only one of them is evidence of what they published.

Not everything that differs is an edit, and saying otherwise would make this archive useless noise. Two things are explicitly excluded:

Site furniture. Navigation, cookie banners and carousels are stripped before anything is compared. A cookie notice appearing between two readings is not a company rewriting its privacy policy.

Our own pipeline. When we change how we capture — switching a page to a rendered browser, or improving the text extractor — the text changes without the publisher touching anything. Those are recorded as recaptures and never counted as edits.

A page that answers with a different company's website is marked served from another domain and is never filed as that company's document. This is not hypothetical: for a period, every vizio.com address answered from outside the United States with an unrelated distributor's site.

Words to accept. For each company, the latest readable capture of every document in its Agreements file is taken, the extracted text is split on whitespace, and the counts are summed. The same text served at more than one address is counted once, by comparing the SHA-256 of the extracted text. Nothing is estimated: the documents are downloadable and the count can be re-run.

Hours to read aloud. The word count divided by words per minute, a deliberately brisk reading pace. A slower pace would make the headline figure larger, which is the wrong direction for a number meant to survive scrutiny.

Edits per month. A capture counts as an edit when its prose differs from the previous readable capture of the same page after site furniture is stripped. Versions recovered from the Internet Archive are dated to archive.org's capture and reported separately from the ones this archive witnessed itself.

Coverage. Documents captured against documents watched, per company and file. A low figure is a statement about this archive's reach, not about the company.

Re-reading an unchanged page costs no storage, because identical bytes hash to a file that already exists. Re-reading a changed one keeps both versions permanently. Removing a URL from the watchlist never removes what was already captured. There is no route in this application that deletes anything.

The full capture record is public: every URL, timestamp, hash, HTTP status and change state, one JSON object per line.

manifest.jsonl Download the archive Live status