Summary

  • The projection generated on 4 November 2026 at 21:40:27 UTC delivered 23 claims for this cable entity: three evidence entries, one service-area mapping, one data-gap summary and eighteen projection or gap records. It delivered zero resource claims and zero relationship claims.
  • Its network layer has never been observed — no ASN, no prefix, no peering relationship, no facility, no exchange — while its identity layer has carried four claims since 17 June 2026. A buyer, a lender or a regulator reading the card as diligence gets silence in the same visual register as substance.

A cable operator whose own record shows no network is either invisible to the instrument or genuinely absent from it, and the instrument cannot tell you which. That is the finding, and it is dated: the projection landed in BTW's directory on 4 November 2026 at 21:40:27 UTC, imported at 21:40:39 UTC, with a confidence score of 0.72. The public card is at Hawaiki Submarine Cable LP Only for Strategic Circle. Everything below is about that card, not about the cable.

Twenty-three claims, and not one of them about a network

The numbers close exactly, which is the first thing to notice. Twenty-three delivered claims decompose into three evidence entries, one service-area claim, one data-gap claim, nine projection claims and nine projection data-gap claims: 3 + 1 + 1 + 9 + 9 = 23. Every one of the twenty-three is partitioned to the anonymous public object card, and no member payload is held behind it. Across twenty-six record-count fields, every counter reads zero. Resource claims: zero. Relationship claims: zero.

So the card is arithmetically complete and evidentially empty at the same time. That is not a bug in the arithmetic; the arithmetic is doing its job. It is a precise statement of what the instrument contains: a great deal of scaffolding about missing material, and no material.

The section-by-section completeness tells the same story without needing interpretation. Identity scores 1.0 and is complete with four claims. Geography scores 0.5 and is partial, missing the country. Event, legal, control, people, contact, network, service and evidence each score 0.0 with a claim count of zero. Nine evidence gaps are named in plain language: no legal claims, no control relationships, no people claims, no public contact claims, no service claims, country, no network resources or relationships, no event claims, no observed or counted evidence.

The operational layer has never once been looked at

Here is the sharpest line in the whole document, and it is a line about time rather than about the cable. The freshness block records a projection time of 4 November 2026 and a latest network observation of null. The network history's first-seen is null. Its last-seen is null. The prefix timeline, the allocation history and the relationship history are all empty arrays.

The identity layer, by contrast, was first seen on 17 June 2026 and has been watched for more than four months. So the instrument has been looking at this entity's name, region, status and confidence for over a hundred days, and in that time it has not recorded a single observation of the layer where an operator actually operates: autonomous systems, prefixes, RPKI origin validation, upstream and downstream relationships, peering, facilities, exchange presence. Zero ASNs. Zero prefixes. Zero total relationships, of every direction and every type. Zero facilities. Zero exchanges. RPKI coverage null, with zero evaluated.

An empty network section is not evidence of a passive asset. It is evidence that nobody has put a network measurement into this record. Those are different facts, and the difference is the whole point of keeping a register separate from a verdict.

Where the card argues with itself, in public

Three contradictions sit on the surface, quotable verbatim.

The first is about what the subject is. One field says: "Hawaiki Submarine Cable LP is recorded as a network operator. Current public evidence covers 2 supporting public references; services, assets, and relationship context should be read with that evidence boundary." Another field, on the same object, says: "Hawaiki Submarine Cable LP appears with 2 supporting public references but no confirmed operator yet; BTW tracks it to link these public clues to a responsible organisation and its relationships." Recorded as a network operator, and no confirmed operator yet. Both sentences are in the payload.

The second is about how much evidence exists. The public-facing lines say two supporting public references. The observability block counts three sources and three source observations, and the evidence list holds three entries. Two is not three.

The third is about geography. A service-area record maps the entity to country "Global" with the stated reason "Hawaiki Submarine Cable LP is mapped to Global for other coverage," while the geography section reports the country as missing at 0.5 completeness. A country called Global is not a country; it is a placeholder wearing the field name of a fact.

None of this is exotic. It is what happens when a delivery pipeline writes prose about its own confidence instead of about the world, and the prose then outlives the confidence.

One address, counted twice, naming another system

The evidence list has three entries. One has no address at all and is labelled "Network directory record." The other two carry the identical URL: https://www.submarinecablemap.com/api/v3/cable/nongsa-changi.json, once labelled submarine_cable_map and once labelled "Public record evidence" at confidence 0.72.

That is one distinct address counted twice. Two entries in an evidence list look like corroboration and are not. And the path of that address names the nongsa-changi system, not Hawaiki.

What the file itself contains, I did not open for this audit, and nothing here characterises it. The finding is narrower and harder: the only external address this record carries, duplicated to look like two, is an address whose own path names a different cable system. Whether that is a mapping accident or a deliberate cross-reference is not established by this payload. Either way, the card's external evidence base reduces to a single address with a naming problem.

The card got thinner, not fuller

Two weeks before this projection, BTW published an inventory of what this record held on 18 October 2026: The cable in the water and the contract that rations it. On that date the record's name appeared inside an owners field alongside BW Digital and Telin; three retained trade reports and one retained release on the adjacent project named Telin, BW Digital and Nongsa Digital Park; and one verbatim passage described the trans-Pacific system in numbers.

The 4 November payload contains no owners field. It contains no retained trade reports and no retained release in its evidence list. Control claims: zero.

Four months after first being seen, the gap did not get filled. The record got thinner. That is the drift worth publishing, because it runs against the assumption that a living directory converges on the truth by accumulation. This one lost an owners field between October and November and gained nothing measurable in exchange.

What our own reporting already established about the cable

So that the audit is not mistaken for agnosticism about the physical plant, here is what BTW published on 24 October 2026 in The cable is fixed. The number is not.: three fibre pairs, a route of roughly 15,000 km, five landing points at Sydney, Mangawhai Heads in New Zealand, Tafuna in American Samoa, Kapolei in Hawaii and Pacific City in Oregon; 43.8 Tbps in the 2018 launch reporting against 67 Tbps on the operator's own network page as retrieved on 24 October 2026; and, in 2022, BW Digital's completed acquisition of Hawaiki Submarine Cable LP together with the announcement of a second system, Hawaiki Nui.

All of those figures are reported here as what BTW published on that date, with that article linked. None of them is freshly verified in this piece, and none of them is an upgrade, a forecast or a utilisation figure. A design capacity is a stock, not a flow; nothing in this record measures traffic.

That distinction matters for the audit. The physical layer of this system is documented well enough for BTW to describe it in public. The operational layer — who announces the routes, who peers where, whose ASN carries the packets — is documented nowhere in the instrument that is supposed to answer exactly that question.

Why an ASN table cannot answer this

A capacity wholesaler and a network operator look identical inside an instrument built from ASNs, prefixes, peering and facilities. A wholesaler need not appear in any of those layers, because its customers can hold them instead: the buyer of an indefeasible right of use may run its own autonomous system over glass it does not own, while the entity that sold the capacity appears in no routing table anywhere.

That is not a loophole in this particular card. It is a structural property of the instrument. Which means an empty network section neither proves nor disproves operation, and a reader who needs the answer gets silence dressed as a gap. The card cannot distinguish "this operator runs no network" from "this operator's customers run the network and the operator sells them the glass."

For a cable system, the question of who controls routing above the glass will be answered by contracts and by the entities at each landing point — not by an ASN table. That is the interpretive boundary this audit insists on, and it is also a practical instruction: stop asking the register to settle questions the register was never built to see.

What would actually settle it

Four documents would move this from gap to finding. An RIR allocation or assignment record, or an RPKI origin, under the limited partnership's own name. A peering-record or routing-registry entry naming an autonomous system the LP operates or is the registrant of. An operator or landing-party document naming the terminating entity at each of the five landing points. And a named legal instrument establishing who may use the capacity and who bears restoration cost when the system is cut.

Any one of those would be a real addition to public knowledge. None of them is in the payload delivered on 4 November 2026. Until one arrives, the honest description of the state of play is that the directory knows the entity's name, its region, its status and its confidence, and knows nothing else about it that a network engineer would recognise as an operating fact.

Who is actually exposed by this

A capacity buyer comparing a private cable against a consortium system cannot use an ASN table to tell which one operates a network, because the seller of glass may appear in none of them. The card looks like it answers the question. It does not.

An investor or lender using a public dossier as a diligence input gets a card whose identity section is complete and whose financials and governance are both marked unresolved. That is the correct label and a terrible surprise if you did not read it.

A regulator deciding whether a landing party is a network operator or a passive asset holder needs an answer this instrument, by design, cannot give.

And the ordinary reader, treating the card as an encyclopaedia entry, is handed a network section that has never been observed at all — rendered in the same typographic register as sections that have been observed. That is where a register becomes a claim by accident: not by lying, but by laying out absence and presence in identical clothes.

The register reads itself honestly if you let it

There is a version of this record that deserves respect, and it is close to the one that shipped. It names nine gaps in plain language. It marks financials and governance unresolved rather than inventing a number. It says, in the data-gap prose, that "network footprint coverage needs stronger reusable public records before this public dossier section can be considered complete." It even tells you what to go and verify: "Verify Hawaiki Submarine Cable LP's legal entity country, current services, visible assets, parent or subsidiary links, and responsible public leadership contacts."

That is an instrument confessing its own boundary, which is more than most directories do. The failure is not that it is empty. The failure is that the confession sits in the same payload as a headline calling the subject a recorded network operator, two count-versus-reality mismatches, a placeholder called Global, and one address counted twice. A reader has to work to find the honesty.

Read the card as a register of what has been observed, never as a statement about what exists. On 4 November 2026 it observed twenty-three things about its own bookkeeping and nothing at all about a network. When a document tells you that much about itself, believe it — and then go ask the contracts.