Summary

  • Internet Exchange Ghana LBG carries a 0.85-confidence directory card derived from one AFRINIC member-list row, which labels it a peering IXP while the card's own legal-type field contradicts the “LBG” in the registered name it reports.
  • A PeeringDB name-substring query returned the verbatim body {"data": [], "meta": {}} on 9 November 2026, and ixpghana.org returned no readable content to two inspection attempts the same day — both are dated readings of a query and a domain, not findings about a switching fabric.
  • The publishable fact is the divergence itself: a registry row, an industry directory and an operator page answer three different questions, and the peer deciding whether to connect is asking a fourth.

The first thing to say about Internet Exchange Ghana LBG is that the record is not silent. It is loud, specific and internally split.

Our own directory card for the entity, last updated on 17 June 2026 (BTW directory: Internet Exchange Ghana LBG Only for Strategic Circle), reports the legal name as “Internet Exchange Ghana LBG” and records one alias: “LINX-Internet Exchange Ghana LBG”. It places the entity in Ghana. It assigns three service facets — managed network, ISP and peering IXP — and marks its own confidence at 0.85. Its stated basis is a single row in the AFRINIC member directory, where the entity appears as a company record for Ghana (AFRINIC membership list).

Then the card contradicts itself. The legal-type field reads private_company. The name ends in LBG, the abbreviation conventionally used in Ghana for a company limited by guarantee. One of those two fields is wrong, and the card does not say which. We are not going to resolve it here; no incorporation document was retrieved, and guessing at a company's legal form from the shape of its name is how directories acquire confident fiction. What matters editorially is that the contradiction sits on the same card as the peering-IXP label, at the same confidence, in the same font.

That card also carries an unusually honest list of what it does not know. Among its stated data gaps: verification of the legal entity country, current services, visible assets, parent or subsidiary links, and responsible public leadership contacts. Read that list twice. A record that says it has not verified current services, while simultaneously assigning three service labels, is not being coy. It is telling you precisely where the label came from — a member-list row with a country and a role, extrapolated into facets — and where it did not.

So register one establishes one thing cleanly: the registry records a member matching this name. Everything above that is card language, not measurement.

The second register is the interconnection industry's own directory, and it said nothing. On 9 November 2026, at 20:28:31 UTC, a PeeringDB exchange-index request for exchanges whose name contains “ghana” returned, in full, this response body: {"data": [], "meta": {}} (PeeringDB exchange index query).

It is tempting to read that empty array as a verdict. It is not. The query was a name-substring match. It can only return an exchange whose indexed name contains the letters g-h-a-n-a in that order. An exchange registered under an abbreviation — GIX is the obvious candidate — or under the alias our own card carries would not match, and the array would come back empty while the fabric hummed along in Accra. We wrote the query; the query's silence is about our string, not about Ghana.

That asymmetry runs in both directions, and it is worth stating plainly because it is the single most common error in infrastructure journalism. Absence from a self-reported exchange directory is a weak negative signal; presence in one is a self-attestation, not an audit. PeeringDB is not an inspectorate. It is a database that exchanges and networks populate about themselves, and its value comes from being current and machine-readable, not from having verified that anyone's ports are lit.

The third register is the operator's own page, and it did not answer. Two separate inspection attempts on 9 November 2026 — at 20:20:13 UTC and again at 20:28:31 UTC — returned no readable content from ixpghana.org (ixpghana.org). No text, no membership list, no port count, no facility address, nothing to quote.

What that means is narrow and should be kept narrow. It is a dated retrieval outcome. A domain can return nothing because it is parked, because it is between rebuilds, because it geo-blocks or bot-blocks automated retrieval, because of a CDN misconfiguration, or because the organisation's public presence lives entirely in a Facebook page and a WhatsApp group, which is not unusual for small African operators. What a failed retrieval cannot do is testify about infrastructure. It does not prove the exchange is dormant. It does not prove it is thriving. It proves that on one day, two attempts, this reader got nothing.

And there is the alias, sitting unexamined at the centre of the card: “LINX-Internet Exchange Ghana LBG”. No retrieved document in this run establishes what that prefix means or where it came from. It may record a brand relationship, a historic sponsorship, a naming convention inherited from a template, or a data-entry artefact from the import. A directory alias is not a routing fact. Nobody should read ownership, governance or operational control into four capital letters that appear in a single field of a record that already contradicts itself on legal form.

The honest summary of the three registers, then: one says a registry lists a member; one says a substring did not match; one says a domain returned nothing. Not one of them is a switch.

What would be a switch? A dated membership list with network names and ASNs. A port count on a named fabric. A route server address that answers a BGP session, or a looking glass showing prefixes learned from more than one peer. A traceroute through an Accra facility that shows a path shortening rather than a hairpin back to Europe. A published peak-traffic figure with a measurement window attached to it. An incorporation record that settles whether this is a company limited by guarantee or a private company, and who its members are.

None of these was retrieved in this run, and it is worth being blunt that the difference between them and what we do hold is the difference between infrastructure and stationery.

Why does the distinction carry money? Because three different readers act on these three registers, and they act differently.

A network engineer in Accra deciding whether to peer is not asking whether the exchange is listed. She is asking whether the round trip to a domestic competitor's content drops from 90 milliseconds via London to two milliseconds across town, and whether the exchange's switching fabric and route server can deliver that. If she reads a peering-IXP facet as an invitation and finds no route server, she has spent engineering hours and a cross-connect fee to learn what the directory never claimed. Her exposure is operational and it is immediate.

An investor pricing West African interconnection reads the same card and sees a national exchange with three service lines and 0.85 confidence. That number is not a probability that the fabric exists; it is our record's confidence in its own derived profile. Treating it as an asset valuation input is a category error with a valuation attached, and it is the kind of error that survives until someone requests a port count in diligence.

A regulator drafting national exchange-point policy reads an empty industry-directory row as evidence of absence and a member-list row as evidence of presence, and both readings are wrong for the same reason: each register was answering its own question. Policy built on either would misallocate a mandate.

The discipline that resolves this is unglamorous. Read the record as a record. Date it. State what was queried, when, and what came back. Show the empty JSON body rather than paraphrasing it into a mood. Show the 0.85 and the data-gaps list rather than smoothing them into authority. Then say what would change the conclusion, so the next reader can go and check.

That is what this article has: a member-list row, a card that disagrees with itself on legal form, an empty array from a name query, and a domain that returned nothing twice in one evening. Ghana's interconnection may be a busy fabric in a downtown Accra building, or a dormant association with a good domain name, or something in between that no public record has caught up with. The evidence we hold cannot choose between those, and pretending otherwise would be the only real error available here.