Step 4 of 4 · Learn to buildHow-to guide

LearnDo it with open toolsRead official guidance as data

Read official guidance as data

An adviser wants an agent to answer from the tax authority's own guidance and cite the paragraph.

A route to listed tools, not a lesson, and not tax advice: a person decides.

Open it

The steps, at a high level

  1. Load a corpus: the French BOFiP as a Hugging Face dataset (8,804 documents with type, validity date, legal identifier and permalink) or the ATO's rulings and guidance as typed CSV and JSON Lines.
  2. Index it for retrieval by document and keep the identifier and permalink with every chunk, so each answer can cite the document it came from.
  3. For EU law, EUR-Lex is the source: free to browse, with a web service for registered accounts; cite the CELEX number.

Check before you rely on it

  • A snapshot has a date: the BOFiP dataset is dated December 2023 and the ATO card warns documents change between captures.
  • Guidance is the authority's reading, not the law; the statute and case law decide.
  • Each answer should quote the paragraph it cites, so a person can check it in one click.

Where these tools sit on the map

Back to step 1The mapWhere tax work happens and by whom: a tax department and a tax authority, team by team, with what is open for each activity.