Summary

  • The operator's own site prints at least four counts of itself on four pages, and every one of them uses a different, undefined or unstated counted noun: 25 member rows, "more than 22 peerings", 24 numbered directory rows and a 13-name association roster.
  • The only traffic figure on those pages is an undated daily "8 Mb" with no measurement window and a unit that cannot be resolved, the membership cost is described as a circuit plus a subscription with no amount, and the exchange that sits across the same city prints the amount with no currency.

On 22 November 2026 the Angolan exchange operated for the members of AAPSI, the country's non-profit ISP association, opened with a claim of "more than 22 peerings" Angola IXP homepage. Its members page listed 25 rows members. Its peerings directory numbered to 24 peerings. Its history page named 13 licensed ISPs history. PeeringDB said 13 peers PeeringDB record.

Five numbers, one exchange, and no two of them the same noun. That is not a hidden scandal and it is not rounding. It is what a record looks like when it prints the count and withholds the definition, page by page, until the only person who can say what the exchange contains is the exchange.

A count on every page

Start with the members page. It prints 25 rows, each with a name, a joined date and an ASN, and no total of any kind. Three of those rows are not member networks at all: "AAI-TEST-IXP" sits at AS65000, and AS37383 appears twice, once as "Angola Internet Exchange Point" and once as "AngolaIXP Route Server". So 25 rows is not 25 networks, and the page never invites the reader to make that mistake. It simply withholds the noun.

The joined dates are worth a second look. Most cluster on 21 January 2026, with the rest spread between 10 October 2025 and 24 February 2026. Nothing on the page states what "Joined" records: a first session, a contract signature, a migration into a new portal. A table where most rows carry one date is at least as consistent with a data migration as with a membership wave, and the record gives a reader nothing to choose between them.

The peerings directory is a different list, not a second view of the first. It is numbered 01 to 24 with columns for organisation, URL, ASN, profile and website, and it carries names the member table does not: Sonangol E.P. at AS36976, Mundo Startel, Movicel, Snet, ITA. It also carries cache entries such as "MicroSoft Cache" and "Google Global Cache" with no ASN in the cell. Row 01 prints its ASN field as "11259 - 37059", two numbers in one column. Row 24 prints "32811", five digits short of the AS328111 the members page assigns to IP WORLD.

The history page supplies a fourth count and a different noun again: 13 INACOM-licensed ISPs named as the association's roster. That 13 belongs to AAPSI's membership, not to the fabric's, and the page wraps it in founding-era prose: created in 2006, a layer-2 platform with a route reflector, local latency of 5 ms to 20 ms against 500 ms to 1000 ms over satellite paths.

Each of those numbers can be true at the same time. That is precisely the problem. A stranger trying to size this exchange has four legitimate readings and no rule for choosing one.

The one number, and everything missing from it

In the six operator pages read for this piece, the history page carries the only traffic figure with a value in it: "Atualmente, o AngolaIXP conta com um volume de 8 Mb de tráfego diário." No date. No indication whether 8 Mb means megabits or megabytes. No statement of whether it is a peak, an average or a daily total, or where on the fabric it is counted.

Everything needed to check the line sits somewhere else. Portal de TI, published 1 August 2025 and crediting Darwin Costa of the Angolan Peering Forum, reported that Angola IXP went above 11 Gbps in the same period in which Angonix reached 59.5 Gbps, and put installed port capacity at more than about 120 Gbps at Angola IXP and about 300 Gbps at Angonix Portal de TI. Treat the site's 8 Mb as 8 Mbit/s and 11 Gbps is 1,375 times it; read it as 8 Mbytes/s and the ratio is 172. Either way the operator's figure sits two orders of magnitude below a number published two years later, and no line on either page says which reading is meant. The division is mine; both inputs are the publishers'.

The exchange does publish graphs. Its statistics page offers Day, Week, Month and Year views and explains that a switch graph shows all traffic handled by a switch, including traffic that originates, terminates and passes through it statistics. In the retrievable page text there is no numeric value, no axis unit and no measurement window; the values are drawn rather than printed.

The clearest sentence on the whole site is about something else. The services page states that the exchange runs several services for public benefit and lists Packet Clearing House, Google Global Cache, Microsoft Edge Cache, an ISC F-root instance, NTP servers and its own statistics services. That is the record stating what is hosted, with no count attached to any of it.

Four registers, four dates

PeeringDB's record prints 13 peers, 17 connections, 107G of total capacity and 64 per cent IPv6, and carries a last-updated field of 13 January 2022. Its peer table holds 17 port or connection rows across those 13 ASNs, with repeated rows for the route servers, ISC F-ROOT LAD1, Paratus Africa and ZAP-Angola. The same record marks the exchange's IX-F member export URL as visibility private.

ISOC Pulse repeats the 13 and says where it came from: members of this IXP, 13 ASNs, are listed in PeeringDB Pulse exchange record. Its 107 Gbps is defined on the same page as the cumulative port speeds of all members, which Pulse calls the potential maximum traffic the exchange could carry at one time, not measured traffic. It also reports one ASN left and none joined in the twelve months before its update, 11 of 13 members using RPKI, and a registrant split of Angola 7, United States 4, Czechia 1, Mauritius 1.

The Euro-IX IXPDB record prints "Number of ASNs 0" and "Number of locations 0" with an update stamp of 19 June 2019 Euro-IX record. Zero in 2019 is not a finding about 2026. It is a record that stopped, and it is printed here as such.

IX Report prints 22 networks across sources, a history row of 18 from IX-F against 13 from PeeringDB for dates from 2 October to 20 November 2026, and a dash in the IX-F column on 21 and 22 November. It also records that the exchange's own website could not be reached when it checked on 17 November IX Report.

Against all of that, the African IXP Association's data export, in rows dated 27 October 2026, carries two Angolan rows, both in Luanda and both marked active, and neither carries a member, ASN or port count Af-IX export. The federation knows the exchange exists. It does not know how big it is either.

Where the publisher's own arithmetic fails

Pulse's Angola country page is the one reviewed surface that adds the two Luanda exchanges together, and it does not close. The page prints "a combined total of 29 members as of November 2026" and, in the same passage, 24 members for Angonix and 13 for Angola IXP Pulse Angola dashboard. Twenty-four plus 13 is 37. The residue is 8, and nothing on the page explains it. Both components are Pulse's; the subtraction is mine.

This is the smallest and most useful failure in the file. A dashboard that adds two self-reported figures and prints a total eight short of its own parts has not told a reader which figure to distrust. It has told them not to add.

The neighbour withholds a different field

Across the same city, the exchange operated by Angola Cables prints what the association's exchange does not: a public-peering price table. A 1GE port at line rate 100 Base-LX carries NRC 500 and MRC 250; a 10GE port at 10000 Base-LR carries NRC 1000 and MRC 500, with the recurring charge described as monthly and the installation charge paid once, and no additional cost for traffic or for use of the route servers Angonix products.

The table prints no currency and no effective date. A reader cannot tell whether 500 is dollars, euros or kwanzas. A 2019 Euro-IX record for the same exchange lists 1G at 950 EUR and 10G at 3850 EUR, an older number in a different currency with no billing period stated. So the two Luanda exchanges fail in complementary ways: one prints the components of a cost and no amount, the other prints the amount and no currency.

The ambiguity migrates across every other surface of the same neighbour. Its own products page lists 38 named networks and ASNs and no port count, no per-member speed and no traffic figure. Its homepage displays "+46,4 Pico (Gb)", a claimed seventh position in Africa, "+23 ISPs" and "8 Anos de Conexão", with no measurement window and no date Angonix homepage. Packet Clearing House's directory row prints 38 participants, a 63G peak, a 33G average, 39,598 prefixes and establishment on 9 March 2015, with no window on the peak or the average PCH directory. Thirty-eight, 38 and 23 are three claims on three surfaces, and the two peak figures, 46.4 Gb and 63G, are not reconciled anywhere.

What no reviewed surface answers

Here is the part that matters for anyone deciding whether to connect, invest or regulate. No reviewed document states how much of Angola's landed international capacity actually transits either exchange. None states how many networks present on the fabric do not peer. None states whether traffic handed over in Luanda replaces international transit or duplicates it.

Three artifacts would close it. A dated port-by-port table from the operator, showing speed, member and session state. An annual report that prints traffic with its measurement period beside it. Or a regulator's interconnection statistics naming the exchange.

Until one of those exists, the record is not silent. It is abundant, and it cannot be recomputed.

Boundaries

  1. The operator's traffic graphs were not read: the exact statistics URL returned a connection timeout and the search-index retrieval carried only headings and the explanatory sentence. The absence of plotted values from that text is not evidence that the rendered graphs lack axes, units or periods.
  2. No reviewed surface prints an Angola IXP traffic figure carrying both a measurement window and a resolvable unit on the same page as a date. The history page's daily "8 Mb" line is the only traffic number on the operator's site, and it carries none of the three.
  3. No reviewed surface prints a port or membership amount for Angola IXP. The history page prints the components of the cost and no amount; no first-party fee schedule was located on 22 November 2026.
  4. PeeringDB's Angola IXP record was last updated on 13 January 2022 and Pulse states that its member count is drawn from PeeringDB; neither is an independent count of the operator's own table.
  5. The Euro-IX IXPDB record for Angola IXP was last updated on 19 June 2019, so its zero ASNs is not a current count and is printed as a dated record, not as a finding about membership.
  6. The IX-F export that IX Report says it read at 00:22 UTC on 21 November 2026 was not obtained directly for this piece; that reading is reported as IX Report's, not as mine, and PeeringDB marks the export link private.
  7. The 11 Gbps figure for July 2025 is trade press quoting a named analyst, not an operator measurement. Its measurement point and window are not stated in the reviewed text.
  8. The rendered membership of the peerings table was only partly recoverable; the visible numbering runs to 24 but the excerpts are incomplete, so no row-by-row reconciliation against the members table was completed.
  9. The Angonix price table prints no currency and no effective date; the amounts are reproduced exactly as printed and no currency is supplied by me.
  10. No reviewed document states why the operator's membership table and its peerings directory carry different populations, or which one the homepage's "more than 22 peerings" counts.