Summary

  • LACNIC's own pages describe routing-security uptake as four differently named quantities — the share of announced routes covered by ROAs, the share of routes in the Valid state, the share of certified resources, and a bare count of hosted ROA objects — and nothing reviewed converts one into another.
  • The 17 September 2025 post prints global IPv4 Valid routes rising from 14.39% to 51.14% and calls that "an increase of more than 400%"; dividing the two printed endpoints gives 3.55 times, or 255%, and the IPv6 pair gives 182% against a printed "more than 900%".

The question a stranger brings to a registry's routing-security page is small and exact: is this prefix, held by this member, protected on this date? What the registry publishes is a regional percentage. Between the two sits a measurement — a numerator, a denominator, a snapshot day, an inclusion rule — and that measurement is the only thing that makes the percentage mean anything.

LACNIC publishes four of them.

The share of announced routes covered by ROAs. The 2024 annual report prints "Tenemos un 54.6% de las rutas IPv4 de la región cubiertas por ROAs y un 55.4% en IPv6" (Reporte anual 2024). The noun is rutas, and the sentence before it names the population: "rutas anunciadas por nuestra región", routes announced by our region. The 2023 report carries the same metric one report earlier at 51.2% and 52.4% (Reporte anual 2023). Two reports, the same wording, the same stated denominator: this is the cleanest pair on any LACNIC page I read, and the movement between them is 3.4 and 3.0 percentage points.

The share of observed routes in the Valid state. A different noun and a different question. The post dated 17 September 2025 reports "Valid routes" — global IPv4 at 51.14% for 2025, IPv6 at 57.01% — with Invalid routes at 6.65% and 6.76% (LACNIC Blog, 17 September 2025). A route "covered by a ROA" is an authorisation that exists; a route in the Valid state is an announcement that matched one. A prefix whose ROA names another origin is covered and invalid in the same second. The post's own figures make the distinction concrete: as the global IPv4 valid share rose, the invalid share rose with it, from 0.48% to 6.65%. The same post reports the "Americas" at 46.10% and 60.87% and states in its own text that the grouping includes North America — so that pair is not the LACNIC service region, and it is not interchangeable with the annual reports' figures.

The share of certified resources. The July 2022 post by the registry's CTO prints "40% de los recursos certificados en la región de América Latina y el Caribe", and later "alrededor del 40%" against FORT monitoring (LACNIC Blog, 14 July 2022). The noun is recursos certificados. No counting unit — addresses, holdings, prefixes — is printed, and no denominator is printed either. The same post is refreshingly blunt about a second confusion: certifying resources and creating ROAs is one goal, operators actually running route-origin validation is another, and the post says the two "generan confusión".

An absolute count of hosted ROA objects. The Tools portal prints four counters in one reading: 16,161 "ASN Registrados", 3,527 "Certificador", 17,699 "ROAs alojados" and 13,522 "Organizaciones" (LACNIC Tools). This is the only figure of the four with no percentage anywhere near it, and the only one that moved while I was reading: a second load of the same URL returned 17,764 hosted ROAs, 65 more, with no printed interval, no as-of date and no stated inclusion rule. "Certificador" is printed but never defined on that page, so I will not tell you it counts certificate authorities, certified members, or anything else.

Only one reviewed page says out loud that the choice of denominator is a choice. An archived LACNIC Labs note from November 2019: "cuando determinamos el porcentaje de cobertura por ROAs de un espacio IP, podemos tomar al menos dos referencias: el espacio total asignado [...] o el espacio total anunciado por BGP" (LACNIC Labs, 27 November 2019). That is the whole published theory of the number. It is six years older than the 2025 post, it accompanies neither annual report, and the same note records that Brazilian members could not yet use RPKI at that date.

Then there is the arithmetic inside one document. The 17 September 2025 post says global IPv4 Valid routes grew "from 14.39% to 51.14%, an increase of more than 400% (in absolute terms)", and IPv6 "from 20.19% to 57.01%", "more than 900%". Divide the endpoints the same post prints: 51.14 ÷ 14.39 = 3.55, a rise of 255%; 57.01 ÷ 20.19 = 2.82, a rise of 182%. Read them as percentage points instead, since the sentence says "in absolute terms", and you get 36.75 and 36.82 points. Neither reading produces 400% or 900%.

Reading the claims onto the Americas pair does not rescue them either: 46.10 ÷ 9.89 = 4.66, and 60.87 ÷ 16.56 = 3.68. The arithmetic is mine; the printed growth figures are the registry's; and they do not reproduce from the printed endpoints. The post is a summary of a study prepared for LACNIC that I did not open, so I cannot say where the 400% came from — only that it is not in the two numbers printed beside it.

Where the record stops, bounded to the seven LACNIC pages I read on 22 November 2026: none prints a count of ROAs that fail validation, that are expired, or that conflict for the same prefix. None records a per-prefix validation state. The Tools portal offers an organisation, ASN and IP query on the captured page — a lookup path for a holder who already knows what to ask, not a published record of outcomes. The institutional service page gives the start dates of the hosted service (1 January 2011) and the delegated service (18 December 2019), and routes Brazilian holders who received Registro.br space to Registro.br's own RPKI page (Certificación de recursos); that routing is not a statement that LACNIC's aggregates exclude that space, and I have not treated it as one.

What this adds: BTW has already examined LAC-2026-3's 80% valid-ROA condition, the 33 certificate authorities in LACNIC's tree converging on one repository path, a Rapport test bench that asks who owns the definition of success, and a route-hijack file. None of them takes the registry's own published figures and asks what a stranger can reconstruct from numerator, denominator, validation status and date. The LAC-2026-3 set already records that the measuring method is to be determined; this is the prior question — what the figures that would feed such a method can currently prove. LACNIC's directory entry records the institution; the instrument is on its own pages.

Nothing here alleges that anyone overstated anything. Each figure is honestly named on its own page, and the 2019 note is candid about the ambiguity. What is missing is the connective tissue: a snapshot day on each percentage, an inclusion rule for the counters, counts of ROAs that fail or lapse, and one dated record per prefix. Those four additions would let a stranger test one case instead of carrying away a regional share — and would make a policy threshold built on "coverage" appealable in something other than prose.