Workshop note · Linux, servers, security
How you tell that a value holds
A date that looks plausible and comes from an official source can still be measuring something other than what you take it for. An example from this project — and the rule that came out of it.
- Revised
- 2026-08-05
- Topic field
- Linux, servers, security
- Figures in the text
- read from the database at request time
Platforms supply a date for every track. It sits in a field that looks like a publication date, comes from the operator itself, and is stored in a clean format. You adopt it without thinking twice. Here that would have been a mistake.
It surfaced as a contradiction: for the same recording the national library’s catalogue named one day of publication and the platform another — twenty-two months later. A single outlier would have been a typo. So the cases were counted.
The cross-check
The ISRC identifies 358 recordings unambiguously in both holdings. For each of them two dates exist. If the platform field were a publication date, the two values would have to agree in most cases.
- platform date later
- 225 cases
- identical
- 0 cases
- platform date earlier
- 2 cases
Not a single match. That is not noise, it is a different quantity: the field measures the day of upload. Two tracks uploaded 43 seconds apart confirm it — a batch operation, not two releases on the same day.
The general part of it: a value is not dependable because it came from an official source, but because you know what it measures. Two independent sources are the cheapest test there is — they cost nothing but the willingness to accept the result when it is inconvenient. That is why each of this network’s 59 recorded sources carries a tier of evidence.
The full procedure is under provenance, the register under sources.