Claims ledger

Each claim, beside what makes it true.

A claim is on this page only if a proof holds it. Each row gives the claim as our pages print it, what makes it true, how you can check it yourself, and the date its proof last passed.

Check my firm What we do not claim

01 · What we claim

The claim, the reason, and your own check.

Every paid assessment is human-verified before release.

What makes it true: The release job holds every paid assessment until a person has cleared it. It has no path to release without that step.
How you can check: Your assessment page shows the date a person verified it.
How we test it: We run the release job on a test record and check that it stops until a person has cleared it.
Reference
agents_page_proofs, G5
Last verified: 8 October 2026
Where we say it: Our AI agents

A finding with no evidence is not released.

What makes it true: A finding stored without its record stops the release, and an approval already given is withdrawn.
How you can check: Every finding in your assessment carries the record behind it.
How we test it: We store a finding with no record, run the release job, and check that it stops.
Reference
agents_page_proofs, G5
Last verified: 8 October 2026
Where we say it: Our AI agents

Every finding is kept with the record that proves it and a check you can run yourself.

What makes it true: Each finding is stored with the record it was read from, the check to repeat, and the date it was read.
How you can check: Run the check beside the finding. It returns the same answer for anyone, your insurer included.
How we test it: We read the stored findings and check that each one holds its record, its check and its date.
Reference
agents_page_proofs, G4
Last verified: 8 October 2026
Where we say it: Our AI agents

A change to your email records is made only when you confirm each step.

What makes it true: The request that writes a record is refused unless it carries your confirmation.
How you can check: Each step you confirm is listed on your dashboard with its date.
How we test it: We send the request that writes a record without a confirmation and check that it is refused.
Reference
agents_page_proofs, G6
Last verified: 8 October 2026
Where we say it: Our AI agents

No token is kept: it is used once and discarded.

What makes it true: The table that records a change has no field for a token, and none is written to any stored row.
How you can check: You can revoke the token at your DNS host as soon as the step is done.
How we test it: We read the table that records a change and check that it has no field that could hold a token.
Reference
agents_page_proofs, G6
Last verified: 8 October 2026
Where we say it: Our AI agents

Our AI agents work from outside only.

What makes it true: The modules that do this work hold no way to sign in anywhere.
How you can check: Every limit is written down in the outside reading policy.
How we test it: We read the code that does this work and check that it holds no way to sign in.
Reference
agents_page_proofs, G8
Last verified: 8 October 2026
Where we say it: Our AI agents

They never log in, never install anything, and never touch your internal systems.

What makes it true: The same modules are searched for any sign-in call, and none is there.
How you can check: Every limit is written down in the outside reading policy.
How we test it: We search that code for every kind of sign-in call and check that none is there.
Reference
agents_page_proofs, G8
Last verified: 8 October 2026
Where we say it: Our AI agents

If the AI fails or reaches its limit, the fixed path runs as it did before.

What makes it true: Each AI step runs behind a switch, a step limit and a spending cap, and when it stops the scheduled checks carry on unchanged.
How you can check: Every finding in your assessment carries the record behind it.
How we test it: We make the AI step fail and reach its spending cap on a test record and check that the same changes go out untouched.
Reference
agents_page_proofs, G15 and stage_agents_proofs
Last verified: 8 October 2026
Where we say it: Our AI agents

The AI writes no finding.

What makes it true: It can only ask for one of our fixed checks on something the first read already listed for your firm, and what is stored is the result of that check.
How you can check: Every finding in your assessment carries the record behind it.
How we test it: We read the code the AI step runs through and check that only a fixed check can write to a record, and that a target not on the firm's list is refused.
Reference
agents_page_proofs, G15 and stage_agents_proofs
Last verified: 8 October 2026
Where we say it: Our AI agents

Every line in those steps is written by us, and AI chooses only where each one goes.

What makes it true: The steps are put together from our fixed steps and the guide we wrote for your DNS host, by a plan that holds positions and no text.
How you can check: Run the check beside the finding. It returns the same answer for anyone, your insurer included.
How we test it: We build steps from a plan and check that every line comes from our fixed steps or the DNS host guide, and that a plan carrying text is refused.
Reference
agents_page_proofs, G15 and stage_agents_proofs
Last verified: 8 October 2026
Where we say it: Our AI agents

It cannot change a severity and cannot hold back a High or Critical change.

What makes it true: It orders the changes the fixed checks found, has no way to write a severity, and a change rated High or Critical always stays in the list to send.
How you can check: Each alert states the change as the check recorded it.
How we test it: We read the Monitor step's code and check that it writes no severity and that a High or Critical change is never held.
Reference
agents_page_proofs, G15 and stage_agents_proofs
Last verified: 8 October 2026
Where we say it: Our AI agents

No firm is named.

What makes it true: The stored figures hold counts only, and the table has no field for a firm.
How you can check: Read the index: each figure carries its sample size and its date.
How we test it: We build the index from test firms with known answers and check that no stored row or printed figure names one.
Reference
leaktrace_index_proofs, X5
Last verified: 8 October 2026
Where we say it: The LeakTrace index

A figure that would rest on fewer than 30 firms is not published.

What makes it true: Small cells are dropped before the figures are stored.
How you can check: Every sample size printed on the index is 30 or more.
How we test it: We build the index from test firms and check that a figure resting on fewer than 30 is dropped before it is stored.
Reference
leaktrace_index_proofs, X4
Last verified: 8 October 2026
Where we say it: The LeakTrace index

Nothing is left off: what is open is listed as open.

What makes it true: The page prints the list the outside reading builds for the same stored record, and has no way to leave a finding out.
How you can check: Read our own reading: it is dated and taken again every day.
How we test it: We store a test reading and check that the page prints every finding the reading builds for that record.
Reference
own_reading_proofs, R2
Last verified: 8 October 2026
Where we say it: Our own reading

02 · What we do not claim

Five things we do not say.

Each of these is kept off our pages by a rule in the same proofs.

That anything is fixed without your approval.
That we act inside your systems.
That anything is watched in real time: checks run on stated schedules.
A promise that nothing gets through.
A count of customers, or a reference from one.

03 · How this page is kept true

It is built from the proofs.

Each claim here is copied word for word from the list a proof holds, and a row whose words no longer match is not shown.
A row is shown only while the page that makes the claim is being served.
The date beside a claim is the day its proof last passed, written down by the command that runs the proofs. It is never the day you open this page.
Where no run has been written down, the row says not yet recorded.