Summary

  • In the participant register I read, an operator approved on 20 February 2026 sits at 0% in the Action 4 RPKI column and is still listed. The rule that could suspend it needs two failing limbs over three months, and no published row says whether that test was ever applied.
  • MANRS publishes its front door twice, in two units: the join page asks for at least one Expected Action across the majority of an applicant's infrastructure, while the specification asks for at least 50% of its ASNs.
  • Section 7 promises that a suspended participant "will be indicated as such on the MANRS website". The table's header carries an approval date, an ASN and five action columns — and no disposition field at all.

The instrument is MANRS: Mutually Agreed Norms for Routing Security, the routing-hygiene badge that networks put in procurement questionnaires and peering conversations. The body that answers for it moved. The Internet Society — the organisation in our directory — still pays for it.

Most instruments I read ask when a party must speak. This one asks who wrote the pass, in what unit, and whether the pass can ever be revoked in public.

One row, four different numbers

Start with what the register actually prints. The header sequence on the participant page is exactly this: "Organization Approved Areas served ASNs Action 1 Filtering Action 2 Anti-spoofing Action 3 Coordination Action 4 IRR Action 4 RPKI". Each row then carries values that are not the same kind of value.

Helikon, approved 27 October 2026, area TR, ASN 198733, reads "Yes 100% 100% 100% 100%". VSX Networks, approved 6 November 2026, area BR, ASN 262355, reads the same way. MREN, approved 30 June 2026, area MW, ASN 37215, reads "Yes –No 100% 100% 100%". SNRD Networks Limited, approved 18 June 2026, areas GB, KR, US, ASN 209710, reads "Yes 100% 100% 100% 93%". Youssef Hamed, approved 31 August 2026, area EG, ASN 207740, reads "Yes –No 100% 100% 80%". University of Illinois, approved 20 February 2026, area US, ASN 38, reads "Yes No data 100% 100% 0%".

So Action 1 is a tick rendered as the word "Yes". Actions 2 through 4 are percentages, a dash-prefixed negative, or the words "No data". A single participant row mixes a binary attestation, three or four measurements and a gap.

That 0% is not, by itself, a scandal. Read the suspension criterion beside it. For Action 4, the actions text sets suspension on failing both limbs over three months — not registering at least 90% of advertised prefixes in an IRR, and not holding valid ROAs for at least 90% of the prefixes a network is authorised to originate. The Illinois row shows 100% in the IRR column and 0% in the RPKI column. One failing limb of two. On the printed arithmetic, that row is consistent with a rule that has not fired.

Which is precisely the problem. The number in the cell is not the number the rule runs on. The rule runs on a two-limb conjunction measured over a quarter. The cell shows one column, with no window, no denominator and no author printed anywhere I could read.

The door is published twice, in two units

The join page states the bar a network must clear: "You (or your company) support the MANRS Principles and implement at least one of the Expected Actions for the majority of your infrastructure." It then adds that implemented Actions "are marked with a check box on the Participants page", and that Coordination "cannot be the only box checked; another Action must also be implemented."

The specification states a different bar in a different unit: "A network operator may become a MANRS participant if at least 50% of its ASNs are MANRS conformant." It also allows relief — "there can be legitimate reasons as to why some requirements cannot be met 100%, and exceptions may be granted where an adequate explanation is provided."

One test is a share of infrastructure. The other is a share of ASNs. Coordination is disqualifying alone on one page and appears as a scored column on the register. Two documents from the same secretariat, two admission tests, no sentence reconciling them.

Then the ongoing test changes unit a third time, to prefixes. Action 4 requires continuing to "register at least 90% of the IP prefixes they advertise" and to hold "valid Route Origination Authorization (ROAs) for at least 90% of IP prefixes or sets of prefixes they are legitimately authorised to originate."

ASNs at the door, prefixes in the corridor, and a bare percentage in the row.

What the percentage is a percentage of

There is a page that defines a per-action measurement, and it does not close the gap. The measurement framework says: "To measure MANRS readiness for a particular network a set of metrics has been proposed, one for each action." For Filtering it counts routing incidents in which a network was implicated as culprit or accomplice, weighted by duration, and produces "a MANRS readiness index (MR-index) for Action1 for a specified period of time."

An index is not a percentage. And the readiness bands for Filtering are printed as "≤1.5: Ready", "1.5−5: Aspiring" and "≥5: Lagging" — a scale with no zero percent and no hundred percent in it. The length of the "specified period of time" is not given in the text I read. Nor does that text mention the participant table, or state that the table's percentages are Observatory measurements.

So the join between the declaration and the measurement is a manual one. A reader takes the ASN from the row, carries it to the Observatory, and reads readiness there — for a network, in an index, on a scale. The Observatory's own description is broad: it "Measures the level of MANRS Readiness of all networks participating in Internet routing, which also serves as an indication of the state of routing security." It also says: "When an operator joins MANRS several checks are performed to ensure that the Actions are implemented."

Several checks. Which checks, on what evidence, kept where, is not printed on the page I read. That is the design question a register of this kind has to answer, and the register does not answer it.

The promise, the body, and the missing field

Now the disposition. Section 7 is unusually direct: "A network operator that has been suspended as a MANRS participant will be indicated as such on the MANRS website and may not participate in discussions relating to the MANRS Actions." Deletion follows a path — "If a network operator fails to undertake the advised actions within 6 months, the MANRS Advisory Group may decide to delete them from the list of MANRS participants" — and a deleted participant "may not re-apply to become a MANRS participant for a minimum period of 12 months."

That is a real mechanism, written down. It is also, on the page I read, an undertaking without an instrument. The register's header has an "Approved" column. It has no column that could hold a suspension, a re-review date, or the date the row was last checked. A promise to mark a suspension on the website is not the same instrument as a column that can hold the mark.

The governance page sharpens the same oddity. The URL about/advisory-group is served under the title "Steering Committee", and the body describes itself as an appeal body "in respect of Secretariat decisions on the outcome of the audit process and suspension and termination of organizations from MANRS participation that fail to meet the minimum conformance criteria." Its one publication sentence concerns process, not outcomes: it "May develop and agree on procedures for specific cases that, once ratified, will be publicly available."

The history page explains how the body underneath all of this changed: "The Internet Society supported MANRS until January 2024, when the Global Cyber Alliance took on secretariat and operational functions. The Internet Society continues to provide funding, advocacy, and training via a grant."

That is the sharpest sentence in the file. The organisation whose name still travels with routing hygiene funds the badge and no longer keeps the register. No page I read gives the funder a conformance role. The funder is a signature on a grant, and the pass is written somewhere else.

What this adds, and where the record stops

BTW has read MANRS twice recently, and neither piece read this. One covered a RIPE 93 proposal to raise the RPKI requirement and the authorisation problem for legacy and leased prefixes; the other counted affected participants in the IRR-filtering minutes, first near 40% and then at 13% in a later detail. Both read a proposed tightening and its arithmetic. This reads the front door, the printed row and the promise.

Across the nine MANRS documents I retained on 19 November 2026, none states what a participant-table percentage is computed from, over what window, or by whom; and none states that a suspension or termination decision is published. The specification's conformance and failure sections reached me in the page's indexed text rather than a complete open of the page, and the versioned PDF of the same actions text — the v2.5.2 file — was not opened; its wording, including the explicit "OR" between the two ongoing 90% limbs, is a search-index excerpt of that file. Where the HTML presents those limbs as consecutive requirements, the PDF excerpt prints an "OR". Whether that connector survives a full read is the one thing in this piece I cannot settle.

A register that prints a number per participant should print, in the same row or beside the table, the unit the number is a percentage of, the window it covers, the body that computed it and the date the row was last reviewed. It should carry a disposition field, because a marker nobody can find is not a marker. And it should reconcile its two entry tests in one sentence, or point the recruitment page at the specification instead of restating it in a different unit.

That is a recommendation about an instrument, not an allegation about a network. The rows I read are internally consistent. The question is whether anyone outside the secretariat could ever prove that.