What we publish, and what we refuse to
This register makes claims about named companies. That obliges us to state the standard those claims are held to, in public, before anyone has to ask.
Effective 3 August 2026
Every claim traces to a document
We publish documented events only. Each statement in a case maps to a specific publication or official document, and that source is linked from the case page. A sentence that cannot be traced to a source is removed from the summary automatically before publication.
Source links are never behind our paywall. Paid products cover convenience and format, never access to the evidence.
Verification standard
An item is published only when all of the following hold at once.
- Two or more independent domains report the same key facts. Syndicated copies of one text count as one source.
- At least one source is tier 1 or tier 2. Tier 3 material is never sufficient on its own, at any volume.
- For cases naming a company where the severity is serious injury or death: two sources minimum with at least one at tier 1. The defamation risk is highest here, so the requirement is strictest here.
- Sources do not contradict each other on company, model, date, place or severity.
- tier 1
- Wire services, national outlets, regulators, court filings.
- tier 2
- Regional and trade press.
- tier 3
- Blogs, forums, social posts.
Automation, disclosed
Cases are assembled by an automated pipeline that uses AI for verification, under the rules above, and publishes without a human approval step. We state this plainly because a reader is entitled to know how the thing in front of them was made.
Two AI passes run independently and do not see each other's conclusions. One establishes whether sources describe the same event and agree. The other looks only for reasons to refuse: confused companies, an old event dressed as new, satire, a single source in several wrappers, named private individuals, accusations without attribution. Anything the second pass flags as blocking stops publication regardless of the first.
Published cases are re-checked automatically after 24 and 72 hours, and their source links are monitored daily. A case that stops clearing the standard is withdrawn without waiting for anyone to notice. Errors are corrected in public.
Fact and metric are not the same thing
The register records facts: an event happened, a regulator acted, a claim was filed, an amount was paid.
The escalation index is a computed metric with an open formula: the share of a company's recorded cases that reached a regulator or beyond. It describes outcomes and press attention together. It is not an assessment of product safety, and we do not present it as one anywhere on this site.
What we do not publish
- Personal data of people harmed. Names appear only where the person spoke publicly themselves or where the name is in open court documents. Names are stripped from headlines and summaries automatically.
- Rumour. An unverified social post is never a basis for a case, only possible supporting material alongside qualifying sources.
- Accusations without attribution. A claim of wrongdoing is published in attributed form or not at all: what a regulator found, what a filing alleges, what a company admitted.
- Speculation about cause. If the cause is undetermined, the case says it is undetermined.
Contact
Corrections: corrections@themachinerecord.com
Company responses: the right of reply form, or reply@themachinerecord.com