Summary

  • RFC 7908 prints six labelled route-leak types and, in its own section 4, states that Types 1 through 4 differ only in "the context of the leaked-route source AS and destination AS roles".
  • RFC 9234 declares a route a leak when the Only-to-Customer attribute arrives from a Customer or Route Server Client; in the sections inspected on 21 November 2026 it prints no Type number and names neither of the two forms that are not role violations.

The sentence that files six types into two groups

RFC 7908 is titled "Problem Definition and Classification of BGP Route Leaks", dated June 2016, and its header prints "Category: Informational". The status sentence on that header is unambiguous about what a reader may do with the text: "This document is not an Internet Standards Track specification; it is published for informational purposes." (RFC 7908, RFC Editor text)

The classification appears beneath a heading that states its own evidentiary basis — "Classification of Route Leaks Based on Documented Events" — and the document introduces it with a limit it never retracts: "This document proposes the following taxonomy to cover several types of observed route leaks while acknowledging that the list is not meant to be exhaustive." (RFC Editor record for RFC 7908)

Six types follow, each with a label a reader can cite. Type 1: Hairpin Turn with Full Prefix. Type 2: Lateral ISP-ISP-ISP Leak. Type 3: Leak of Transit-Provider Prefixes to Peer. Type 4: Leak of Peer Prefixes to Transit Provider. Type 5: Prefix Re-origination with Data Path to Legitimate Origin. Type 6: Accidental Leak of Internal Prefixes and More-Specific Prefixes.

Then comes the sentence that does the filing. "It is worth noting that Types 1 through 4 are similar in that a route is leaked in violation of policy in each case, but what varies is the context of the leaked-route source AS and destination AS roles." (RFC Editor record for RFC 7908) Four of six, told apart only by who sits at each end of the announcement.

The two that are not filed that way turn on what the prefix is rather than who its neighbours are. Type 5: "A multihomed AS learns a route from one upstream ISP and announces the prefix to another upstream ISP as if it is being originated by it (i.e., strips the received AS path and re-originates the prefix)." Type 6: "An offending AS simply leaks its internal prefixes to one or more of its transit-provider ASes and/or ISP peers" — internal prefixes, the document adds, often more-specific prefixes covered by an already announced less-specific one.

Section 4 also anticipates the overlap and declines to multiply the taxonomy for it: "A Type 5 route leak (i.e., prefix mis-origination with data path to legitimate origin) can also happen in conjunction with the AS relationship contexts in Types 2, 3, and 4", and then, "Hence, the special cases mentioned here are not included in enumerating route-leak types." A reader can see the discipline: one type per mechanism, role context handled as context, not as new species.

The document also tells you where the measurements live instead of making any of its own: "For measurements of valley-free violations in Internet routing, see [Anwar], [Giotsas], and [Wijchers]." (RFC 7908, RFC Editor text)

What the later document actually tests

RFC 9234 is titled "Route Leak Prevention and Detection Using Roles in UPDATE and OPEN Messages", dated May 2022, header "Category: Standards Track". Its status sentence is the opposite kind of licence: "This is an Internet Standards Track document." (RFC 9234, RFC Editor text)

The mechanism is a negotiation, not a policy table. The document "uses a new configuration parameter, BGP Role, which is negotiated using a BGP Role Capability in the OPEN message", and "An optional, transitive BGP Path Attribute, called 'Only to Customer (OTC)', is specified in Section 5".

Five roles are defined for an external session — Provider, Customer, Route Server, Route Server Client, Peer — each with its own propagation permission, and the document states plainly that "Violation of the route propagation rules listed above may result in route leaks [RFC7908]." It also does the translation work a stranger would otherwise have to do: "The terms 'Provider', 'Customer', and 'Peer' used here are synonymous to the terms 'transit provider', 'customer', and 'lateral (i.e., non-transit) peer', respectively, used in [RFC7908]."

The trigger is a role test, stated as such. "If a route with the OTC Attribute is received from a Customer or an RS-Client, then it is a route leak and MUST be considered ineligible." From a Peer, a value that does not match that Peer's AS number is likewise a leak. The attribute itself is "an optional transitive Path Attribute of the UPDATE message with Attribute Type Code 35 and a length of 4 octets", and its purpose is to enforce that a route sent to a Customer, a Peer or an RS-Client "will subsequently go only to the Customers". The procedures apply only to AFI 1 and AFI 2 with SAFI 1 (unicast) and must not be applied elsewhere by default. (RFC Editor record for RFC 9234)

The document names one boundary and prints it as a prohibition rather than an omission: "Roles MUST NOT be configured on an eBGP session with a Complex peering relationship" — Complex meaning the local AS holds more than one role toward the remote AS. It is candid about what that costs: per-prefix policy can approximate the same outcome but provides "no in-band measures to check the correctness of the per-prefix peering configuration".

What the inspected sections do not print

Here is the finding, and it is a negative one, so it stays pinned to what was actually read on 21 November 2026: the RFC Editor's rendered text of Sections 1 through 8 of RFC 9234, plus its reference entry for RFC 7908.

Across those sections, no Type 1 through Type 6 label appears. The only use of the word "Type" is the attribute code. No sentence in them mentions prefix re-origination, internal prefixes, or more-specific prefixes. No sentence in them declares any route-leak form residual, uncovered or out of scope; the single "outside the scope" sentence in Section 5 is about deployment details for private AS numbers behind an Internet-facing ASN. And [RFC7908] sits in Section 9.1 Normative References, cited for the definition of a route leak and for those Provider/Customer/Peer synonyms — the entry itself enumerates no types. (RFC Editor record for RFC 9234; RFC Editor text)

Now the join, which is mine and not either document's: the mechanism's trigger is exactly the role-context family that section 4 of the taxonomy describes, and the two forms whose definitions turn on what the prefix is rather than who its neighbours are appear neither in the trigger nor in any exclusion sentence. If you want the two prefix-keyed forms named beside the mechanism, you compose that sentence yourself, holding two documents side by side.

That is not a charge against either text. A taxonomy that says its list is not exhaustive has not promised anyone a mechanism per type, and a mechanism that restricts itself to unicast IPv4 and IPv6 has printed its own perimeter. The observation is narrower and, for anyone who has to explain routing incidents to a board, more useful: the classification and the detection test in this record do not share a vocabulary of types at all. One speaks in labels, the other in roles, and the bridge between them is a reader's work.

Two status sentences, two different jobs

Both documents print the sentence that governs what a reader may do with them, and both sentences matter more here than usual. The taxonomy is "published for informational purposes" — it is the document you cite to say what a route leak is and what forms it takes, and it is comfortable admitting the list is open. The mechanism is "an Internet Standards Track document" — it is the document you implement, and its permissions are MUSTs and MUST NOTs on a session's role configuration.

Stacked in that order, the pair reads sensibly. A problem defined widely in 2016; a mechanism scoped narrowly in 2022 to what can be negotiated in-band between two neighbours; the interval between June 2016 and May 2022, five years and eleven months by my subtraction, spent on the negotiable part. The gap in vocabulary is the price of that focus, and it was paid in public.

One name, five and five

The taxonomic document's header prints its authors in order: K. Sriram; D. Montgomery; D. McPherson; E. Osterweil; B. Dickson, with "US NIST" and "Verisign, Inc." as affiliations. (RFC 7908, RFC Editor text) The IETF Datatracker record prints the same five names, dates the document June 2016, traces it to draft-ietf-grow-route-leak-problem-definition in the GROW working group, and names Joel Jaeggli as the responsible Area Director. (Datatracker record for RFC 7908)

The mechanism's header prints: A. Azimov (Qrator Labs & Yandex); E. Bogomazov (Qrator Labs); R. Bush (IIJ & Arrcus); K. Patel (Arrcus); K. Sriram (USA NIST). (RFC 9234, RFC Editor text) Its Datatracker record dates it May 2022, calls it a Proposed Standard, traces it to draft-ietf-idr-bgp-open-policy in the IDR working group, and names Alvaro Retana as responsible Area Director. (Datatracker record for RFC 9234)

Five names and five names, and the intersection is one: K. Sriram. The RFC Editor's authors' addresses block for the mechanism expands those initials — Kotikalapudi Sriram, at the National Institute of Standards and Technology in Gaithersburg, Maryland. (RFC Editor record for RFC 9234) The 2016 document prints only initials and its affiliation block, no correspondence address; I read that header, not a back-matter page for it.

What a byline can and cannot carry

Neither document attributes a sentence, a section, or a scope decision to any individual author. Nothing in either text says who drafted section 4, who proposed the role test, or why the mechanism stopped where it did. Anyone who tells you otherwise is reading the byline as testimony.

What the byline does carry is real, if modest: the same organisation's name sits on both, and one individual's name sits on both, across two working groups, two Area Directors, and two publication categories. That is the whole of the person-level claim, and it is enough for the point this record makes. The taxonomy and the detection test were not written by strangers in separate worlds; they were written by overlapping communities, in the same standards process, one with a name in common on the file.

And the name is doing no persuasion. If you swapped that single common name for any other, the two documents would still print the same six labels, the same role negotiation, the same OTC attribute, the same absence of a type-mapping sentence. What the swap test tells you is the honest limit of a byline piece: it identifies who was in the room for both conversations, not who chose the threshold.

What the pair is good for

For an operator, the practical upshot is unromantic and useful. If you configure a role on a session and exchange OTC, you have bought detection for the leak forms that announce a route to the wrong class of neighbour — the four the taxonomy groups by role context. You have not bought a detector for a multihomed network stripping an AS path and re-originating a prefix, or for a transient internal-prefix spill. Those remain, in the taxonomy's own terms, forms defined by the prefix rather than by the relationship, and no sentence in the inspected mechanism text promises otherwise.

The uncertainty that actually bears on the claim: Sections 6 through 9 of RFC 9234 were read through the RFC Editor's rendered text and the Datatracker record, not in one uninterrupted pass from the plain-text file, and the reference entry is the only part of Section 9 I examined. Sections 5 and 6 of RFC 7908 and its back matter were not read. No implementation report, deployment measurement or operator survey of OTC was read for this piece — so this is a reading of two documents, not a statement about how much of the internet is now covered by either.