How LTC works

It finds the moment. You make the decision.

LTC runs when a site asks you to agree to something. Here is exactly what happens, in order.

1. Detection, on your device

A small piece of LTC runs on the pages you visit. It watches for the shapes a consent request takes: a cookie banner, a signup form with a terms checkbox, a subscription screen, an app-permission prompt, an OAuth consent screen, a wallet approval.

It reads the page only to classify that moment. It looks at the accessible names of the buttons, whether anything is pre-ticked, how big each control is, how many steps declining takes, and which of the site’s own policy pages the screen links to. It does not read the article you were reading, the message you were writing, or anything you have typed into a form.

2. Nothing is sent until you say so

LTC ships with remote analysis off. In that state it works entirely on your device: it assesses how the choice is presented, and whether the access being asked for is proportionate to what the service does. It cannot read the company’s agreements, so it never claims to have.

When you turn remote analysis on, what leaves your device is the site’s domain, which kind of consent is being asked for, counts and sizes of the controls, and links to the site’s own published policies. Not the page. Not the address you are on. Not anything you type. There is no field in the request that could carry them.

3. Working out which documents apply

A consent screen usually links to the agreements that govern it, and that is the strongest possible signal: the site itself is telling LTC which terms apply. Where it does not, LTC falls back to what it already knows about that company.

LTC only retrieves documents a company has published for the public, and it respects robots.txt. It does not get past a login, a paywall, or any access control.

4. Reading the clauses

LTC splits the document into passages and matches each against a versioned taxonomy of 56 clause categories — data sale, AI training, automatic renewal, mandatory arbitration, and so on. Most of this is deterministic pattern work, not a model: a rule cannot invent a clause that is not there.

Where a passage denies something, LTC records the denial rather than the claim. “We do not sell your personal information” is a commitment, and reporting it as a data sale would be the most damaging mistake this product could make.

5. Evidence, or nothing

Every material thing LTC tells you is attached to a character range in a specific version of a specific document. Before anything is shown, LTC re-checks that the passage really is at those offsets in the document it stored. A finding that fails that check is discarded, not softened.

6. What it shows you

  1. What could hurt you — worst first, always.
  2. What you get — real protections, found the same way.
  3. If you decline — with the basis LTC believes it on, and “LTC does not know” when that is the honest answer.
  4. LTC view — a suggestion. Never an instruction.

What LTC will not do

  • It will not press a button for you, or accept or reject anything on your behalf.
  • It will not tell you something is illegal. It will tell you what a clause says and, separately, what a named rule provides.
  • It will not say “safe”. An absence of findings is an absence of findings.
  • It will not stay quiet when it fails. “Unable to verify” is a state you will see.