How regulatory monitoring works
Texts evolve continuously. The Normivia catalog has to keep up, without ever integrating an update without human review.
1. Automated monitoring, every month
Three sources are queried: EUR-Lex for European law, Légifrance for French law, and private bodies (SWIFT, ISO...) for their own standards. Each source is handled independently: a failure in one pass never prevents the others from running.
2. Two types of detection
The "watchlist" monitors standards already tracked by Normivia, looking for amending acts. "Discovery" looks for entirely new texts not yet present in the catalog.
3. AI-assisted summary, only for genuinely new candidates
A summary is only generated for a detection that isn't already known, never recalculated on every run for an already-identified candidate.
4. Human review before any update
Every detection goes through a review queue: a Normivia expert explicitly validates or rejects it before the regulatory catalog changes. Nothing is ever integrated automatically.
A status that's always explicit
Never a false silenceWhen a source isn't configured (for example, missing Légifrance access), the result is logged as such, never confused with "ran but found nothing". The difference stays visible to the team at all times.
Source: backend/veille_pipeline.py, backend/veille_store.py