A derived database is a reading of somebody else's documents. Readings contain mistakes — parser bugs, changed formats, ambiguous fields. The only defence that works is making every claim checkable, and then saying so.
What identifies a filing
An accession number identifies one submission uniquely and permanently. Given it, the original document can be retrieved from the regulator's own system and compared line by line with what we show. The number appears on the records that came from a filing.
The filer is identified by a CIK, which does not change when a company renames itself or changes ticker. If a company page looks like the wrong company, the CIK settles it.
What identifies a security
Fund holdings are reported by security identifier rather than by company. Where we could not link that identifier to a company page with confidence, the issuer name is shown exactly as filed and left unlinked, because a wrong link is a false statement while a missing link is only an inconvenience.
What identifies a list entry
Sanctions entries carry the authority's own identifier and belong to a specific published version of the list. Every search result states which snapshot it came from and the date that snapshot was published, so the answer can be reproduced or refuted against the authority's current file.
What identifies a rate
Each published rate carries the date it was published for, and every conversion states which date it used. A converted amount without a date is not checkable and, on this site, is not produced.
When we are wrong
If the original document and our page disagree, the document wins. That is not a courtesy — it is the only workable rule for a system that reads other people's files at scale.
The changelog records source connections, section changes and corrections, so that a figure that changed can be traced to the change that caused it rather than appearing to have always been that way.