Summary

  • LACNIC's published record for AS14624 names PSINet do Brasil S.A., tax identifier 03.085.145/0001-20, as registrant, administrative contact and abuse contact; the registrant entity was last changed on 7 June 2001.
  • The routing plane last saw AS14624 originate a prefix on 17 September 2001; at the 19 November 2026 query, zero of 318 RIS peers saw it in IPv4 and zero of 310 in IPv6.
  • The Brazilian operation was reported closed in 2001 after its sale processes failed. The insolvency proceeding published in Rio de Janeiro names PSINet do Brasil Ltda., not the S.A. in the registry, and no document read bridges the two.

A number with a name on it

Somewhere in the public registration data that Brazilian network operators query every day there is an object numbered 14624. It begins where it ends: startAutnum 14624, endAutnum 14624. Its name is ASN14624, its type is DIRECT ALLOCATION, its country is BR. The entity standing behind it — registrant, administrative contact and abuse contact in one — is PSINet do Brasil S.A., tax identifier 03.085.145/0001-20 https://rdap.lacnic.net/rdap/autnum/AS14624.

That is the whole of what the record asserts, and it is worth being exact about what it does not. It does not say the network is running. It does not say the company still answers the address in its own contact fields. It does not say anyone trades, employs anyone, or has paid a bill this century. It says that for the purposes of the numbering layer, this number is held by that legal person. Everything else about AS14624 — the traffic, the customers, the circuits, the insolvency — belongs to other ledgers, and the registry is not one of them.

That separation is why the record can look careless and still be right. A registry is a uniqueness-coordination instrument: one holder of record per number, so that a dispute has an address to be served on. It is not a network monitoring system, and it was never asked to be. The moment you read it as a statement about reachability you have changed the question it answers.

Three clocks in one record

Read the timestamps and the object splits into three vintages. The aut-num object itself carries a registration of 2013-04-01T18:10:20Z and a last change of 2013-04-01T18:10:58Z, thirty-eight seconds apart — the signature of something being created rather than maintained. The registrant entity was registered on 2000-02-10T12:00:00Z and last changed on 2001-06-07T14:05:54Z. The contact entity was registered on 2001-06-06T12:00:00Z and last changed on 2001-08-23T22:22:10Z https://rdap.lacnic.net/rdap/autnum/AS14624.

The subtraction is mine, not the record's: those autnum stamps sit more than eleven years after the registrant's last change. Read 1 April 2013 as an allocation date and you would be claiming that a Brazilian internet provider registered its network handle a year and a half after the São Paulo press reported the operation shut. The 2013 event belongs to the object; the 2000 and 2001 events belong to the entities. A record can be re-rendered in a database without its holder lifting a finger.

The more interesting dates are the older ones. The registrant entity was last touched six days after PSINet Inc. announced its Chapter 11 filing. The contact entity was last touched three days after the update line of 20 August 2001 on the Estadão report that the Brazilian operation had closed. Nobody in the material read explains either change, and no causal link is claimed here. But the sequence is the whole story in miniature: this record was still being edited while the company whose name it carries was being dismantled, and then it stopped.

The last prefix anyone saw

The routing evidence is unusually clean at both ends. The first observation attributed to the origin is prefix 204.6.98.0/24 on 2000-08-18T08:00:00. The last is prefix 200.188.81.0/24 on 2001-09-17T08:00:00. At the query time of 2026-11-19T16:00:00 the same response prints zero visible prefixes, zero addresses, zero observed neighbours, zero of 318 RIS peers seeing the AS in IPv4 and zero of 310 in IPv6 https://stat.ripe.net/data/routing-status/data.json?resource=AS14624. The overview endpoint for the same AS prints the holder string “AS14624 - PSINet do Brasil S.A.” and announced false at the identical query time https://stat.ripe.net/data/as-overview/data.json?resource=AS14624.

Two dates twenty-five years and two months apart, and no continuous series between them in the material read here. That gap is a boundary, not a finding. What the snapshots support is narrow and sufficient: at the far end of the period, no sampled vantage point could see this origin announcing anything.

A network that stops being announced has not been deleted. Announcement is a claim about who is using the number right now; registration is a claim about who holds it. The two ledgers can diverge by decades and neither becomes false. And the counting rule matters: zero of 318 peers is a statement about visibility from a fixed set of observation points, not a census of the internet, and not a legal status. Writing “the number is unused” from that zero would be the same category error in the other direction.

Two endpoints, one record

Query the American registry's RDAP service for autnum 14624 and you get AS14624 as a DIRECT ALLOCATION in BR, registrant PSINet do Brasil S.A., the same tax identifier, and a self link pointing at rdap.registro.br https://rdap.arin.net/registry/autnum/14624. Two endpoints, one record. This matters because a diligent reader who checks both will conclude that two institutions agree. What actually happened is that one institution's record was served twice.

A mirror is its own evidence class and it inherits nothing from the registry it mirrors. It is useful for reach — more people will stumble on the American surface than on the Brazilian one — and useless for corroboration. Where this article relies on the Brazilian record, it relies on the record, not on the number of places the record can be read.

What the parent's filing did not do

PSINet Inc.'s own announcement on 1 June 2001 said that the company and 24 U.S. operating subsidiaries had filed Chapter 11 in the Southern District of New York, and that four Canadian subsidiaries had filed under the CCAA in the Ontario Superior Court of Justice. In the same document it said the Asian, European and Latin American operating subsidiaries, and Metamor, were not part of the filings https://www.sec.gov/Archives/edgar/data/940716/000090130901500044/exh99_jun12001.txt.

So the most quoted fact about PSINet — the bankruptcy — does not by itself reach the Brazilian company whose name is in the registry. That is not a claim that Brazil was unaffected; the parent was trying to sell it. It is the narrower and more useful point that the filing is not the document that decides the Brazilian entity's fate, and treating it as one is how a plausible chain of reasoning acquires a missing link.

The same release lists what was already moving: a letter of intent with TELUS for the Canadian operations and facilities, a definitive stock purchase agreement to sell the Panama operations to REE Panama, S.A., a closed sale of substantially all of the Puerto Rico operations, and Latin American alternatives under discussion with a purchaser group and no assurance of a sale https://www.sec.gov/Archives/edgar/data/940716/000090130901500044/exh99_jun12001.txt. The parent was not hiding the problem. It was telling the market, in the announcement of its own collapse, that Brazil was unresolved.

The buyer's later filing dates the same voluntary petition 31 May 2001 and describes PSINet as debtor-in-possession from 31 May 2001 through 1 April 2002 https://www.sec.gov/Archives/edgar/data/1158324/000091205702024458/a2082455z8-ka.txt. One petition, two printed dates a day apart: the company said “today” on 1 June, the buyer's filing says 31 May. Both are in the file and neither is corrected here. A one-day difference between an announcement and a docket entry is the ordinary friction between a press release and a court record, and it is not evidence of anything.

Two failed sales and a closed network

Brazil's fate was settled inside a sale process that failed twice. On 7 June 2001 the trade press reported that an investment group including senior PSINet Latin American managers had signed a letter of intent to buy the operations in Argentina, Brazil, Mexico and Uruguay, and noted both that the Latin American subsidiaries were not part of the 1 June filing covering 24 U.S. operating subsidiaries and that the petition claimed liabilities of US$4.3 billion against assets of US$2.2 billion https://www.internetnews.com/archive/psinet-considers-offer-for-latin-american-operations/. A letter of intent is not a sale, and the numbers are the petition's own claimed totals as reported.

By 2 October 2001 La Nación reported that the joint sale to Cori Capital Partners had failed, that the parent had then given each country operation freedom to act, and that Brazil closed because no buyer could be found, while Chile and Panama found buyers https://www.lanacion.com.ar/economia/netizen-comprara-la-cartera-de-psinet-nid339837/. The Brazilian closure itself is Estadão's story, updated on 20 August 2001: PSINet do Brasil ended its operations after negotiations to sell all or part of the business failed, the parent having also failed to sell the Latin American asset group including Brazil; the operation had been the country's third-largest provider after 11 acquisitions, had more than 3,000 corporate customers by June, and advised customers to subcontract local companies https://www.estadao.com.br/link/psinet-encerra-suas-operacoes-no-brasil/.

Two details on that page are worth printing rather than smoothing. Its furniture shows a later display timestamp with an update line of 20/08/2001, so the original publication date is not cleanly represented; and it uses “concordata” for a United States Chapter 11, which is not the Brazilian legal term for that proceeding. Both push the correct reading: a contemporaneous newspaper account of a closure, not a court record of one.

A fifth account complicates the sequence usefully. On 20 June 2001 the Washington Post described a company selling itself piecemeal after the filing — a deal worth roughly US$77 million for the Canadian operations, a buy-back of a software business — and reported that PSINet had earlier found buyers for most of its Latin American operations https://www.washingtonpost.com/archive/business/2001/06/21/psinet-continues-selling-operations/4032174d-713d-424b-8f05-9d79bdbabd5c/. That sits awkwardly beside an October report of a failed group sale and a Brazilian closure. It is not a contradiction: a buyer can be found and a sale can still fail to close, and the two reports are four months apart in a process that was re-pricing weekly. In date order, the sequence is the finding: interested buyers in June, a group sale dead by October, one country shut for want of a purchaser.

What the buyer bought

The American end is documented to the asset. The asset purchase agreement is dated 26 February 2002; the closing was 2 April 2002 for approximately US$9.5 million in cash; the enumerated assets were accounts receivable, equipment, certain fibre IRUs, customer contracts, certain intellectual property, the PSINet tradename, selected peering arrangements, and leases for three web-hosting data centres https://www.sec.gov/Archives/edgar/data/1158324/000091205702024458/a2082455z8-ka.txt. The buyer's own press release announces completion of the acquisition of the majority of PSINet's U.S. operations after court approval on 27 March 2002, describes the assets as the customer base, the backbone network and associated equipment, and rights to intellectual property, and states the price as US$10 million https://www.cogentco.com/en/about-cogent/press-releases/279-cogent-communications-acquires-us-operations-of-psinet.

Two prices for one transaction, four months and one audience apart, both from the buyer. Neither corrects the other: a filed figure carries a filing's precision and a release carries a release's rounding. Both are printed here because a reader who meets only one of them will think the other is a discrepancy.

The part that belongs to this story is what the enumeration does not contain. The tradename moved. The list of acquired assets names no number resource — no autonomous system number, no address block. That is a boundary of a summary description, not proof that nothing transferred: the filings say nothing about AS14624 in either direction, and the deal was U.S. operations. But it does mean the most detailed public account of the break-up of PSINet's American network cannot be read as evidence that anyone took the Brazilian number.

The Ltda. in the filing

The Brazilian insolvency trail runs through a differently named company. A court-appointed administrator's page lists a PSINet do Brasil Ltda. bankruptcy proceeding, process 0112333-53.2001.8.19.0001, with a bankruptcy decree dated 19 December 2001 in the 6th Business Court of the Capital of Rio de Janeiro https://cmm.com.br/processo/falencia_psinet-do-brasil-ltda-__317. Ltda. and S.A. are two different corporate forms in Brazilian law, and nothing in the material read on 20 November 2026 connects the Ltda. in that proceeding to the S.A. in the registry: no corporate-chain document, no register extract, no filing that names both.

This is the joint in the story that cannot be closed by inference. A decree against a Ltda. does not dissolve an S.A.; an entry naming an S.A. does not prove the S.A. is alive. Two readings fit the evidence equally well: a group with more than one Brazilian vehicle, one of which went through insolvency; or a single business whose legal form changed somewhere between the registry entry and the court file. No document in this material chooses between them, and the original court papers were not obtained.

Anyone who tells you the registry entry is stale because the company was declared bankrupt is skipping that step. The registry names one legal person; the decree names another. The gap is the accountability question, not the answer to it.

What a dead brand does in a routing table

Nothing. BGP distributes numbers, not corporate identity, and no field in the protocol has room for a parent company's bankruptcy, a tradename sale or a change of corporate form. That is why the routing plane went quiet in September 2001 without registering an event: an origin stopped appearing, and the protocol's business was finished.

In the registry the name does more consequential work. It is the answer to “who holds this number?”, which is the only question the layer exists to answer, and here it points at a single entity occupying the registrant, administrative and abuse roles at once https://rdap.lacnic.net/rdap/autnum/AS14624. Whether that channel is still reachable is not established by the material read on 20 November 2026. The real cost of a twenty-five-year-old entry is not that the name is wrong. It is that nobody has verified the address behind it and the record has no field in which to say so.

BTW's own directory entry for PSINet do Brasil S.A. points at the same AS14624 https://btw.media/en/directory/psinet-do-brasil-s-a. So does the holder string in the routing overview, which prints the company's name and the false announcement flag in one response https://stat.ripe.net/data/as-overview/data.json?resource=AS14624. Three surfaces, one name, and a network nobody has seen since 17 September 2001 https://stat.ripe.net/data/routing-status/data.json?resource=AS14624.

Old is not stale

Every registry record ages. The question is not whether this one is old — it is, by a quarter of a century — but whether the layer that holds it has been told something better. On the evidence read here it has not. The registrant's last change is 7 June 2001, the contact's is 23 August 2001, and the object's own event is a 2013 system stamp that says nothing about the holder.

A record is stale when reality has moved and nobody has corrected the record. A record is merely old when nobody has told the registry that reality moved. The difference decides who carries the burden of proof. The party claiming the registry is wrong has to produce the dissolution, the transfer or the successor; the registry does not have to re-prove itself every morning. That is not indulgence of bad data. It is the only rule that makes a public register stable enough to be relied on at all, because a register that treats silence as vacancy would let every patient reader of an old entry become an owner.

The record also enforces a second discipline: registration, authorisation and announcement are three ledgers, and an empty announcement count closes exactly one of them. AS14624 is registered to a named holder. It is not announced from any sampled vantage point. Neither fact says who may use the number, and neither says what happened to the address space once originated under it. Turning the zero into “the number is free” is a category error, and it is precisely the error that converts a dormant record into somebody's opportunity.

What would make the record wrong

Four things, and none of them is in the material read.

A registry action. If the Brazilian registry had reclaimed, transferred or re-issued AS14624, the registrant field would not still read PSINet do Brasil S.A. No such action appears in what was read, and no correspondence explaining the record's silence was obtained either — a gap in the record, stated as a gap.

A corporate-chain document. A register extract, a merger instrument or a filing naming both the Ltda. and the S.A. would settle the entity question in one page. Nothing read does.

An announcement. A prefix originated by AS14624 tomorrow would not make the registry wrong. It would make the two ledgers diverge, which is the accountability event worth watching, because the party operating the number would not be the party the registry names.

A reply. An authenticated response from the administrative or abuse contact would change no date at all, and it would retire the only genuinely worrying thing about a record this old.

What is in hand instead is an entry that has outlived its network by a quarter of a century, a corporate history that ended in failed sales and a closed operation, and an entity mismatch the file does not resolve. The registry has not been caught being wrong here. It has been caught being old, which is a different accusation, and whoever wants to make it stick still has to file the paper.