We publish machine-produced analyses of contested questions, and those analyses name real organizations and real people. Sometimes we will get something wrong. This page says what to do about it.

Two different kinds of page carry an Arbiter verdict, and this policy treats them differently.

Reports we publish — the research reports on this site — are ours. We commissioned them, we published them, and we answer for them.

Investigations published by users at /share/… are theirs: their sources, their reliability scores, their conclusions, published on their own say-so. We host those pages. We do not write them, review them before publication, or stand behind them.

Reports we publish

Report an error

Write to corrections@housecarl.cloud.

Tell us the page, the specific statement you believe is wrong, and — if you can — what the correct position is and where it is documented. A link to a primary source is the fastest possible route to a fix.

You do not need to be the subject of the report to file a correction. If you are the subject, say so; those go to the front of the queue.

What we do

Acknowledge within 72 hours. A human reads it and replies. Not a form-letter autoresponder.

Resolve within 30 days, usually much sooner. One of four outcomes:

1. Correct it. We fix the error and the page carries a dated correction notice saying what changed. We do not silently edit and pretend the original never happened.
2. Annotate it. Where the disagreement is about interpretation rather than fact, we add your position to the page, attributed to you.
3. Withdraw it. If a report rests on a source that turns out to be wrong, the conclusion does not survive the source. We unpublish.
4. Explain why not. If we think the report is right, we say so and show our reasoning. You are entitled to disagree publicly and we will link to your response.

What counts as an error

We measure sources against what they say. We do not adjudicate the world, and we will not pretend to. That makes the scope of a correction narrow and specific — three categories we always act on, and one we do not.

Misrepresentation of a source. A misquoted figure, a claim attributed to the wrong source, a study described as saying something it does not say, a date or a name that is wrong. This is our error in the strictest sense: we said a source said something, and it did not. We fix these.

Miscomputation. If the fusion, the reliability weighting, or the temporal decay was computed wrongly, the verdict is wrong on our own terms. We fix these and say what changed.

Evidence about a source's reliability. A retraction, a correction issued by the source itself, or credible evidence that a source fabricated or materially misstated its data. This is not a disagreement with us — it is input to us, about the one thing we claim to measure. A system that weighs sources by reliability cannot coherently refuse to update when handed proof of unreliability. We re-run the investigation with the new evidence weighted in and republish with a dated correction notice, or withdraw the report if the conclusion does not survive its sources.

Disagreement with a verdict is not, by itself, an error. A verdict of "unsupported" means the cited evidence does not carry the claim — not that the claim is false. But if you think a source is missing, or a weighting is indefensible, tell us. A missing source is a real defect and we would like to know about it.

Investigations published by users

A page at /share/… was written and published by an Arbiter user. We did not commission it, we do not review share pages before publication, and hosting one is not an endorsement of it.

Share pages are open to search engines and may appear in search results. That is a deliberate change from our earlier posture, in which they were marked for exclusion.

If you are named in one, this is what taking a page down actually means, in order of how fast each part happens. It is removed from our database immediately. Published pages are also held in an edge cache to make them load quickly, and that copy expires within one hour of removal. A search engine then drops it on its own schedule, which we do not control and cannot accelerate. If a page about you is still appearing in search results after we have removed it, tell us and we will confirm in writing that it is gone from our end — search engines have a removal request form for exactly this situation, and that confirmation is what they ask for.

If a share page concerns you, write to corrections@housecarl.cloud with the page address and what you say is wrong with it.

We acknowledge within 72 hours, the same as anything else.

We pass the complaint to the person who published it. They wrote it. They are the one who can correct it, and they are the one you have a dispute with.

We will take the page down. A share page is a courtesy we extend to the person who made it, not a right they hold against a third party. Where a complaint is specific and made in good faith, we unpublish and tell the publisher we have done so. The published snapshot is deleted and that investigation cannot be re-shared afterwards. They remain free to publish elsewhere, and we take no view on whether they should.

We will not adjudicate the dispute. We do not rule on whether the publisher was right about you, and taking a page down is not a finding that they were wrong. Refereeing a dispute between two other parties is not something we are equipped to do, and pretending otherwise would do you no favours.

Complaints about copyright are handled through a separate process with its own statutory requirements. Write to the same address and say that is what you are sending.

Why we do it this way

A published verdict about a named organization is worth nothing if we are not willing to be wrong about it in public. A correction is cheaper than an argument and considerably cheaper than a lawsuit, and a house that corrects quickly is one you can trust the rest of the time.

That is a promise we can make about our own work. We cannot make it about someone else's, which is why the two halves of this page read differently.