Methodology

Routes

Once a day, every RIPE RIS and RouteViews collector's full-table snapshot from 08:00 UTC (a collector without one uses its latest earlier snapshot that day). Each is reduced to one row per prefix and origin AS, with the peers that see it. A peer (address + AS) feeding several collectors counts once. A full-table peer sends at least 90% of the median full table of its address family; visibility is how many of them see a route, and a route seen by fewer than 10% is labelled low visibility. A network is an origin AS: one with at least one route has a page. A route whose AS path ends in an AS_SET has no single origin and is on no network's page.

RPKI

Our own rpki-client run with the five RIRs' trust anchors, the last run before the snapshot (at most 4 hours before; none means RPKI is unavailable that day). Route origin validation as in RFC 6811:

Valid: A ROA for the route's origin covers it and its length is within the ROA's maxLength.
Invalid: maxLength: A ROA for its origin covers it, but the route is longer than that ROA's maxLength. RIPEstat calls this "invalid length" also when the covering ROAs are for other origins; here that is "invalid: origin", so the fix it points to is the right one.
Invalid: origin: ROAs cover it, none for its origin (an AS0 ROA, the holder's statement that the space isn't to be routed, counts here and is named in the fix list).
No ROA: No ROA covers it.

IRR

The day's dumps of the five RIR databases and the other public ones RADB mirrors (RADB, ALTDB, NTTCOM, JPIRR, RIPE-NONAUTH and others; LEVEL3 and IDNIC publish no dump and aren't read, so a route object only there counts as missing). The RIR databases are authoritative: they check who holds the space; the others don't.

RIR database: A route object with the route's exact prefix and origin is in an RIR database.
Other IRR only: Such an object is only in non-RIR databases.
Other origin only: Route objects for the prefix name other origins.
None: No route object for the prefix. When a route object for a shorter prefix with the same origin covers it, the page says which.

Route objects not announced: objects with a network's AS as origin whose prefix no peer sees that AS announce. Many are kept on purpose (backup routes, space announced only in some places).

AS-SETs

From each network's irr_as_set in PeeringDB. Expanded from the day's dumps the way bgpq4 does against whois.radb.net: a name with a database (RIPE::AS-FOO) is read from that database only, members included; one without from the first database that has it in RADB's order (RADB, LACNIC, IDNIC, AFRINIC, RIPE, RIPE-NONAUTH, …). Members are followed to the end; sets that list each other are noted as a loop. Route-sets and single AS numbers aren't expanded. Checked against whois.radb.net's own expansion of large sets: the same within 0.2% (our dumps are up to a day older).

Route servers

For each exchange member on the route servers (IX Report's lists): how many of the routes it originates a route server would leave out if it dropped RPKI-invalid routes and filtered on the IRR the way IXP route servers commonly build their prefix lists: the member's AS-SETs from PeeringDB (with none usable, its own AS number), expanded as above, with more-specifics of a route object for the same origin accepted down to /24 and /48 (bgpq4 -A -R 24 / -R 48). Only the member's own routes are counted: the collectors show each route's origin, not the network it came through, so its customers' routes aren't in these numbers.

Bogons

Prefixes in a block of IANA's IPv4 or IPv6 special-purpose registry that isn't globally reachable, IPv4 multicast, IPv6 outside 2000::/3, and default routes; origins that IANA's AS number registries don't list as assigned to an RIR (private use, documentation, AS_TRANS, unallocated). Blocks the registry marks "N/A" (TEREDO, 6to4) aren't counted.

Fix list

The changes that would bring RPKI, the IRR and what's announced into line, one row each: ROAs to create or whose maxLength to change, route objects to add (in the RIR database for the space) or to remove, AS-SET entries. They are facts about the data; whether a change is wanted is the network's call.

What the data can't show

The collectors' peers are a sample of the Internet: a route none of them receives isn't here, and a route seen by few may be seen widely elsewhere.
One snapshot a day: routes that come and go between snapshots aren't seen.
Only the origin of each route is used, not the path, so a network's customers' routes aren't counted as its own.
IRR dumps can be up to a day older than the live databases; PeeringDB and the exchanges' member lists are what networks and exchanges publish.

The MANRS Observatory, RPKI monitors and bgp.he.net cover parts of this. What's here besides: every network, the exchange view, and RPKI, the IRR and AS-SETs checked against each other. Sister sites: IX Report (who is on which exchange) and RIR Tracker (the registries' changes to address space and AS numbers).