You should be able to read back what you signed.

The problem

A waiver is handed over at a counter, on a tablet, in a queue, with a child already in a jump suit. It is signed in under a minute and rarely seen again. There is no copy in your email, no account to log into, and no version number. The terms are not secret — most are published somewhere — but they are scattered, long, and written for a court rather than a customer.

The moment that matters is usually later. Somebody gets hurt, a membership will not cancel, a photograph turns up in an advertisement — and the question becomes what the document actually said on the day it was signed. By then the business has often reworded it, and the page that held the old text is gone. That is the gap this index exists to close.

What we do

Four things, in order:

• Capture. A document is collected from its published source and stored with the date it was captured and a hash of what was captured, so a record can be checked against the original.
• Read. Clauses worth reading twice are picked out and described in plain language, each one quoting the paragraph it came from.
• Track. When a document changes, the old version stays. The index shows what moved between them.
• Publish. Everything is searchable and linkable, by company, by location, by clause type, and by state.

The rule the whole thing runs on

An explanation never replaces the document. Every plain-English reading sits on its own surface, says what produced it, and quotes the exact paragraph it is based on. If the two disagree, the document is what matters — so the document is always one click away, never summarised out of view.

Worth being plain about, because most sites like this one would not be: no model reads these documents. A keyword pass decides which of six clause types a document contains and records the paragraph that triggered it; each clause type has one fixed sentence describing what that kind of provision generally does. The attention score is the sum of a fixed weight per clause type. That is the whole of the analysis. It is a smaller claim than a generated summary would be, and it is one we can stand behind line by line.

Three voices carry that rule through every screen:

• Interface and explanation — the sans-serif you are reading. Anything we wrote sits on a tinted surface and says where it came from.
• Original wording — a serif on warm paper. Reproduced, never paraphrased.
• Provenance — monospace. Capture dates, hashes, section refs, versions, counts.

What we don’t do

We do not tell you whether a clause is enforceable, whether a document is legal, or whether you should sign it. Those are questions for a lawyer who knows your situation and your state. The attention score on a document reflects how much of it a reader may want to read carefully — it is not a rating of the business, and a high score does not mean a document is unfair.

Who it’s for

Anyone working out what they already agreed to: a parent after an injury, reading what was signed on a child’s behalf; a member trying to find the cancellation window; somebody who has just been told their claim goes to arbitration and wants to see the clause that says so. Researchers and journalists looking at how a provision spread through an industry. Businesses that would rather their terms be understood than skimmed. Everyone sees the same record.

Where the data comes from

Businesses that collect waivers electronically host the signing page on their vendor’s platform and link to it from their own site. Those links are public, and the Common Crawl archive has followed them — so the index is built by replaying pages that archive already holds, rather than by crawling anybody’s server. Every document then gets fetched once from its own page and compared against the archived copy, which is what a verified badge means here: collected from the published source, and re-checked since. How we verify sources sets out what each status means.

Nothing is guessed at. A capture is only published when the wording names the business on two independent signals — because a record filed under the wrong company is worse than no record. Documents that do not clear that bar stay out of the index and are counted below.