Summary

  • The RouteViews decks of 30 May and 19 June 2024 each print a report date, a count of "BGP routing table entries examined" and a count of "Total ASes present in the Internet Routing Table" — and neither prints a time of day, a collector or peer count, an address-family scope sentence, or a written definition of either noun.
  • Twenty days separate the two printed dates; the entry count falls by 155 while the AS count rises by 27, and nothing on either page lets a stranger decide whether the internet moved or the measuring did.

A deck filed for PacNOG 33 opens with the title "The Internet Routing Table Using RouteViews" and, in its opening text, the presenter line naming Philip Smith. Inside it sits an IPv4 report headed "IPv4 Routing Report 19th June 2024", printing "BGP routing table entries examined: 952989" and "Total ASes present in the Internet Routing Table: 75930" (the PacNOG 33 deck). The contact address printed beside the name is not reproduced here.

Twenty days earlier, a second deck in the same archive prints "IPv4 Routing Report 30th May 2024", "BGP routing table entries examined: 953144" and "Total ASes present in the Internet Routing Table: 75903" (the ThaiNOG 6 deck). Its retrieved opening text shows the title and no presenter line, so the attribution rests on the archive rather than the page: the RouteViews presentation index lists that item under 31 May 2024 as a presentation by Dr. Philip Smith of NSRC, and lists the PacNOG 33 item the same way (the RouteViews presentation index). Both files sit in one directory listing (the /presos/ index).

My subtraction, not the decks': twenty days separate the printed dates. Between them the entry count falls by 155 (953,144 to 952,989) while the AS count rises by 27 (75,903 to 75,930).

Two counters, two directions

A reader holding only these two pages has no way to choose between the innocent explanations and the uncomfortable ones. A table can lose entries to aggregation while gaining the networks that announce them. A different set of feeds can add a hundred and fifty entries without anything on the internet changing at all. Either would produce this pair of numbers, and so would a genuine reversal in one direction and growth in the other.

The decks do not help. There is no time of day attached to either measurement, so the interval between them is twenty days wide and an unknown number of hours deep. There is no count of collectors or peers, so a reader cannot tell whether the same instrument produced both readings. There is no sentence scoping the report to IPv4 or to any address family, though the heading says IPv4 and the entry showing 952,989 is the one a reader would take as the whole table.

And there is no prose definition of either counted noun anywhere in the retrieved text of either deck — read on 21 and 22 November 2026, from the exact URLs above, as extracted page text rather than as opened slide images.

That last gap is the one that matters most, because a count is only checkable against a definition of what was counted. "Entries examined" could mean unique routing entries, entries after some aggregation, or something the reader is expected to already know. Without the sentence, 952,989 is not a measurement anyone outside the room can re-run. It is a number carried by a person's standing.

The page that does define it, and the reason not to use it

A definition exists in the same software family. The background page for the Global R&E Network Routing Report states that "BGP Routing Table Entries Examined: Lists the number of unique routing entries in the Internet Routing Table", and describes a report fed by organisations volunteering an EBGP multihop feed into a collector hosted by NSRC at the University of Oregon (the NSRC background page). That page belongs to a different report. Lifting its sentence onto these decks would be an inference wearing a citation, and the decks would still not have said what they counted.

The same page does print the cadence the decks leave out: "The report is produced on a weekly basis, with the mail shot based on data captured at 4am Australian Eastern Standard Time (+10GMT) every Saturday." It is the wrong document for the definition and the right one for showing that the sentence can be written at all.

The sibling that prints the clock

The weekly IPv4 Routing Table Report is a stranger to that silence. Its 17 October 2026 edition opens with "IPv4 Routing Table Report 04:00 +10GMT Sat 17 Oct, 2026", names its vantage point as "IPv4 BGP Table (Global) as seen at Japan", and then prints "BGP routing table entries examined: 1083516", "Total ASes present in the Internet Routing Table: 79532", "Number of addresses announced to Internet: 3127057152" and "Percentage of available address space announced: 84.5" (the 17 October 2026 edition). The edition of 4 April 2026 prints the same shape: 04:00 +10GMT, 1,042,729 entries, 78,243 ASes, 3,108,788,352 addresses announced, 84.0 per cent (the 4 April 2026 edition). Between those two printed dates, 196 days by my subtraction, the entry count rises by 40,787.

A time and a zone, a named vantage point, labelled nouns: three of the four things the RouteViews decks withhold. The fourth — a printed definition — is missing here too. And the name attached to this mailing is a contact line, not a byline: the 1 September 2007 edition carries Philip Smith's name in its comment line, alongside the same 04:00 +10GMT stamp and the same labelled counts (the 2007 archive). A contact line tells you who answers mail about the number. It does not tell you who computed it, and I am not treating it as though it does.

Nineteen years of a percentage without its base

Every edition of this mailing prints a percentage of available address space announced and, on the line above, the number of addresses announced. None prints the number the percentage was divided by. Do the division yourself, across three editions nineteen years apart: 1,756,836,704 divided by 0.474; 3,108,788,352 divided by 0.840; 3,127,057,152 divided by 0.845. All three land within about seven million addresses of 3.70 billion. That is my arithmetic, not the report's, and its steadiness is the point: the figure is computed against a fixed base that the report never names.

A reader who wants to know whether 84.5 per cent is a lot, a little, or a constant in disguise has to reverse-engineer the denominator.

The contrast sits in the same lineage. An archived copy of the CIDR Report dated 9 May 2008 prints a table headed "ASnum NetsNow NetsAggr NetGain % Gain Description" and a row reading "Table 265937 167031 98906 37.2% All ASes" — a percentage with both of its inputs in the same row, on the same page (the 2008 copy). The live front page now credits "Original Concept: Tony Bates, Revised by: Philip Smith, Further Revised: Geoff Huston" and prints a generation stamp and two columns, Date and CIDR Aggregated (cidr-report.org today). The current maintainer's own account of that succession, published on 24 March 2026, is plainer still: "The CIDR Report was first produced by Tony Bates, then taken on by Philip Smith, who then passed the baton to me" (the APNIC Blog account).

So the family knows how to print a base. It printed one in 2008 and it stopped. That is a choice about pages rather than an impossibility, and it is the same choice the RouteViews decks make when they print a noun without its meaning.

The annual report that is not his

One document in circulation resembles these decks closely enough to be mistaken for them. "BGP in 2025" on the APNIC Blog is bylined to Geoff Huston on 8 January 2026, and it prints a collection frame — 777 BGP sessions, 398 of them providing an IPv4 table (BGP in 2025). It is not Philip Smith's instrument, and this piece does not attribute it to him. It is named here because it shows what a printed frame looks like in the same corner of the field, and because a reader who goes looking for an annual series under Smith's name will find this one first.

Where the arithmetic has to stop

Three fields cannot be checked from what was read. The time of day behind either deck's measurement is not printed. The collector and peer frame behind either deck is not printed. The definition of the counted noun is not printed on either deck, and the page that does carry a definition belongs to another report. The documents that might close the gap — speaker notes, dropped slides, a definitions page on the report's own site, the full message bodies of the mailing editions beyond the excerpts read, the per-AS sub-pages of the CIDR Report — were not obtained in readable form, and I will not pretend a missing slide is a hidden method.

What remains is exact and enough: two dated pages by one engineer, twenty days apart, printing a count that moves down and a count that moves up, with nothing on either page that would let a stranger say why. Whether 952,989 or 953,144 is the better count of the internet on those two days is not a question these decks let anyone answer.

The deck has a date. It needs a clock, a frame, and a definition — the three sentences its sibling already prints, and the one it used to.