Limits
Known limitations
What the dataset cannot tell you, and why — stated up front rather than discovered mid-investigation.
Collected here in addition to appearing beside the features they affect. An unanticipated limitation is more costly than a documented one.
| Limitation | Consequence |
|---|---|
| IPv6 is not collected by probing | A miss means “not collected”, not “clean”. Some external lists cover it and are named where they answer. |
| ASN and geo enrichment reaches ~41 % | Everything ASN-based is limited to that slice |
| December 2024 and January 2025 are missing | Trends spanning them are not comparable |
| August 2025 is nearly empty | Same |
| Three providers are no longer probed | perfectprivacy, astrillvpn, hotspotshield show history only |
| Mullvad probing has broken down | Three addresses in 30 days against hundreds historically |
| Five collectors, one contributing 21.6 % | The corpus is geographically skewed |
| No network-type data without PeeringDB | Collateral estimates need that mirror synced |
| No ASN-to-organisation grouping | CAIDA’s licence prohibits redistribution, so “shared ASN” cannot be resolved to “shared operator” |