Statistics and freshness

Detection lag is computed as first_seen_at − published_at, in seconds: the gap between the filing timestamp the AMF publishes and the moment our poller first saw the record.

Bulk-imported documents have no measurable latency. Their first_seen_at is NULL, they are excluded from these statistics, and no freshness badge is shown for them. A displayed zero would mean the poller saw the record in the same second it was filed, which does not happen.

This is not our latency and should not be read as such. The AMF export refreshes roughly every 29 minutes and runs a measured 61 minutes behind actual dissemination (see docs/SPEED.md); we poll it every 60 seconds, 29 times more often than the source changes. Polling harder would change nothing — the ceiling is upstream. And uin_dat_amf is the time the issuer declared, not a receipt: 36% of filings carry an exact hour or half-hour against the 3% a uniform spread would give. The figure above measures the gap between a declared time and our detection, not the speed of this site.

Source: info-financiere.gouv.fr, published by the Autorité des marchés financiers under the Etalab 2.0 licence. The figures above describe our copy of the corpus, not the AMF’s complete corpus.