The instrument record as a machine-readable file
The record is published as a file so a compliance reading can be rebuilt from the same sources. Each row is one recorded statement with its origin and its date. As of 2026-09-12.
Legal text moves slowly and commentary about it moves quickly. A file of what the official sources actually say, with dates, is the part that stays useful when the commentary has moved on.
Rows come from published instruments and from vendor pages, and the two are never merged. A duty recorded from an official text and a product claim recorded from a marketing page are different kinds of statement, and the source column keeps them apart.
Where an obligation has a date of application, that date is in its own column rather than buried in the wording. Dates are the part of this subject that most often gets quoted wrong.
The licence below allows reuse, including in commercial compliance work, provided the attribution stays attached. A clause quoted without its source is not evidence of anything.
1What each column holds
| Column | What it holds |
|---|---|
| fact_id | A reference that stays fixed while the text around it is amended. |
| value | The article text or vendor statement, quoted. |
| applies_to | The paragraph or product the row records. |
| source_url | The published text it was read from. |
| source_name | The title that source carries. |
| value_since | The date the wording is known to hold from. |
| verified | When the source was last consulted. |
Inclusion rule. Everything recorded is exported, including the rows that record silence in a vendor's pages. Order. File order follows recording order, which groups rows by instrument and then by vendor.
The file is at /datasets/clauses.csv, licensed ODC-By 1.0. Attribution should name the record and the reading date.
Elsewhere in the record: EU, Duties, Generators, Statements, Pairs, Fields, Questions, Terms, Learn. How an instrument is read is set out on the sourcing page.