Summary

  • Acosta's bylined LACNIC posts attach a numbered reference to nearly every percentage, and the 26 June 2024 review defines the counted noun outright: 30% means 30 of every 100 people already have IPv6.
  • The cited pages print the inputs — per-country end-user curves credited to Google's JSON, or Google's own users' share — not the LACNIC-wide aggregate his sentence states, so the citation travels and the aggregation rule does not.

The definition comes first, before any number does. In the English ten-year review: "The numbers you will see in the following images represent IPv6 penetration rates among end users. For example, 30% means that out of every 100 people, 30 already have IPv6" (LACNIC, 26 June 2024). The Spanish original says it in its own idiom — "de cada 100 personas, 30 ya cuentan con IPv6" (Un viaje de 10 años) — so the denominator is not a translation artifact. The method paragraph is just as plain: the collector is "simply a python3 script that connects to Google Statistics [1], processes and cleans the data, and then stores it in our database."

The output is one sentence, printed twice: "In December 2023, 34% of our region had IPv6," and again in the conclusions, "By the end of 2023, 34% of the region had adopted IPv6."

Now the footnote. Open Google's IPv6 statistics page and it says what its number is: "The graph shows the percentage of users that access Google over IPv6." Google counts Google's users. The metric is labelled IPv6 Adoption, and a provider-reported reading of the same URL printed "Native: 46.59% 6to4/Teredo: 0.00% Total IPv6: 46.59% | Nov 19, 2026". No averaging window, no user count, no claim about the world.

Open LACNIC's own graph-access page and the same thing happens one level down. Its axis legend reads "el grado de penetración de IPv6 en el usuario final"; it charts each country in LACNIC's coverage since May 2014; it credits Google — "Stats based on: https://www.google.com/intl/en/ipv6/statistics.html json file for this stats" — and it stamps itself "Updated daily at 01:58 UYT" and "Page generated by: graph-access.py." What the retrieved text does not print is a LACNIC-wide percentage, a base, a measurement date, or a count of the users behind it.

So the citation is honest about the instrument and silent about the join. Country curves, each a share of Google's users in that country, become one share of "our region" in Acosta's prose; the rule that turns the first into the second appears on neither page reviewed. That is the whole finding, and it is narrower than it sounds. Nothing here says 34% is wrong. A regional aggregate can be perfectly computed and simply never displayed.

My arithmetic, and only mine: 34 minus 30 is 4 percentage points. The 30% is "approximately 30%" under the word "currently" in a 3 January 2023 post (Strengthening IPv6-only capabilities); the 34% is for December 2023, printed in June 2024. Two instruments, two time references, and no sentence anywhere saying the two were computed alike. A four-point gap is something I can subtract; a four-point annual gain for one defined population is not something the reviewed pages establish.

That 2023 sentence is the best citation grammar in the corpus, and it is worth reading closely because it gets the hard part right. "Currently, global IPv6 penetration among end users is over 40% [1], while in the Latin American and Caribbean region it is approximately 30% [2]" — one sentence, two nouns, two instruments, [1] Google's page and [2] LACNIC's. What it does not print is a window. "Currently" dates the sentence, not the measurement.

By 9 June 2025 the noun has moved. "With nearly 50% of global traffic now running over IPv6 (and almost 40% in Latin America and the Caribbean)" (7 challenges IPv6 faced). Traffic is not end users; a share of bytes is not a share of people. The reference list names LACNIC's statistics page, and no separate global source is named in the retrieved text for the nearly 50%. Two shares of two populations are not one series, whatever the shape of the curve between them.

The 2021 ASN post shows the habit at its sharpest, because there the author states the classification rule himself: "A efectos de las estadísticas mostradas se considera un ASN con IPv6 aquellos que reporten más de un 1 por ciento (>1%) en su adopción (valor tomado por APNIC)" (ASNs con IPv6 vs ASNs sin IPv6). The data has for years come from APNIC, he writes, and the chart redraws on Saturdays at 3:59 pm UTC-3.

That parenthetical deserves a fair hearing rather than a verdict. In the APNIC material reviewed for this brief, the map is headed "IPv6 Capable Rate by country (%)" and prints a "30 day average (22/10/2026 - 20/11/2026)" (APNIC IPv6 measurement maps); the AS page is headed "IPv6 Deployment for AS" with an "Average Interval (days)" control set to 1 (AS-level page); the methodology post describes 25M-30M ad impressions a day, defines capability as the proportion of tested end users who can retrieve an IPv6-only object, and weights each economy with UN Statistics Division and ITU-T Internet-user counts (How we measure: IPv6); and an APNIC presentation hosted by ARIN says "44% of the region's economies are now visible in IPv6 capability (above 1%)" (APNIC update, ARIN 51). The one 1% threshold that surfaces governs whether an economy is visible. It does not govern whether an autonomous system counts as IPv6-capable. That is a citation not corroborated in the pages reviewed here — not a misattribution, and certainly not proof that APNIC never wrote the rule. The file that would settle it is the raw ASN dataset the 2021 post links; it did not open during retrieval (read timeout), so its fields, threshold and timestamps stay unverified.

If any of this reads as a charge of carelessness, the corpus itself answers it. In a 2019 LACNIC news item — a piece that quotes him rather than carrying his byline in the retrieved text — Acosta says: "We obtain many of these numbers from our own systems and others from external sources, always crediting our sources" (IPv6 users up 50%). The same item describes a population-weighted regional average built from each country's population, the region's total population and each country's degree of IPv6 penetration. That is the closest thing to an aggregation rule anywhere in this material, and it sits in a 2019 quotation, not beside the 2024 figure.

Where the arithmetic has to stop: the country-level values in the 2024 post exist only inside images; Google's and LACNIC's chart labels render client-side and were absent from the text read; and the APNIC ASN dataset did not load. Nobody holding these pages on 21-22 November 2026 can recompute 34% or nearly 50% for their stated populations. Every quotation above is provider-reported page text extracted from those URLs on those dates, not a rendered first-party reading — and the dynamic pages will print other numbers on another day.

That is why the citation is not the claim. The claim is the aggregation rule, and it belongs on the page where the number is printed: which countries, which population weights, which window, computed by which script. Acosta already writes the definition sentence most analysts skip. The line that would let a stranger divide is the one the corpus still owes the reader.