Summary

  • Read on 12 November 2026, ix.nz set out its facility presence street by street: five Auckland rows in the excerpt retained, with the block breaking off mid-list. The same host named three document channels — a Technical Guide, a Document Library and a Network Status page.
  • Everything retained from that reading carried no route-server address, no statement that the switching path is redundant, and no availability, maintenance-window or outage-handling undertaking. That is a bounded absence in the readable record, not a finding that the operator promises nothing.

The record starts at street level

A facility block on the landing page reads Facility Presence and then lists, verbatim: Datacentre220 at 220 Queen St, Auckland; 2degrees DC at 7a Parkhead Pl, Albany, Auckland; Spark MDR at 31 Airedale Street, Auckland; DataVault at 162 Grafton Road, Auckland; and DataCom – Orbit Data Centre at 6 Orbit Drive, Rosedale, Auckland.

Five rows, all in Auckland, and the excerpt stops mid-block. Five is what the retained text names; it is not a facility total, and nothing here should be read as one. Most of the names on those rows belong to other companies — Spark, 2degrees, DataVault, DataCom, Datacentre220. The exchange sits inside halls it does not own, which is exactly what a carrier-neutral fabric looks like when you write it down.

The same host, in the same reading, describes the operator as not-for-profit and carrier-neutral, places the fabric in Auckland, Wellington and Christchurch, and sets out 10G and 100G peering ports alongside a Chorus EdgeConnect access alternative. Those are page-level characterisations rather than quoted excerpts, and they are reported as such.

This is not a reopening of the earlier finding about exchange volumes. BTW's October reading of these pages established the facility and participation register and the absence of an exchanged-volume figure with a unit, an averaging interval and a scope. That question stays answered where it was answered. The question here is narrower and harder: what does the public record say about how the fabric keeps working?

Three names in a navigation line

The line is short enough to quote whole: Technical Guide Document Library / Network Status.

Read what that actually is. It is navigation text. It tells a stranger that the operator maintains, or intends to maintain, the three document classes that would answer a configuration and recovery question — how to attach, where the reference material lives, and what the fabric is doing right now. It is a signpost, and signposts are useful. A signpost is not the road.

That distinction is where most infrastructure reading goes wrong. A label like Network Status looks like evidence of operational transparency, and in a link audit it will be counted as one. It is evidence of a place to put an answer. Whether the answer is there, whether it has a date, whether it distinguishes a planned maintenance window from an unplanned fault, and whether anyone is on duty to update it at three in the morning — none of that is in the label.

What was retained from the participants page in the same reading was no page text at all, only a characterisation that it lists current participants and publishes no historical total of ports. What was retained from the metrics page was a recorded failure to open, with no text or measurement behind it. Those are retrieval results, and they belong in the notes rather than in the argument.

What the readable record does not carry

Stated as narrowly as it can be: in the text retained from the landing host on 12 November 2026, there is no route-server address, no statement that the switching path is redundant, and no availability, maintenance-window, incident or outage-handling undertaking.

That sentence is doing careful work and it should not be stretched. It does not say the operator promises nothing. It does not say a route server does not exist. It does not say a technical guide is empty — the guide is named, not read. It says that within the material actually retrieved and kept on that date, the configuration and recovery layer was not present in readable form.

Absence in a reading is one of the easiest things in this trade to convert into a claim about the world, and one of the most damaging when the conversion is wrong. The discipline is simple to state and hard to hold: a page that did not open is not an empty page, and a document named in a menu has not been read.

The test I am applying, and its status

Three elements, read together, would let a stranger treat a fabric as a documented layer rather than an adjacency:

  1. A configuration surface. Where a member attaches and how the fabric is addressed, in readable prose or a readable table — not a document title.
  2. A redundancy fact. Whether the switching path survives the loss of a component, stated as a property of the fabric rather than implied by the number of sites.
  3. An availability undertaking. What the operator commits to when the fabric degrades, how a member escalates, and what response is promised.

This rubric is mine. It is a reading discipline, not a published standard and not an operator claim, and the exchange has not failed a benchmark here because no benchmark was applied to it. Naming that plainly matters, because a made-up yardstick dressed as industry practice is exactly the kind of thing that gets quoted back years later as if a regulator had said it.

Why the gap is structural, not scandalous

The asymmetry has a mechanism, and the mechanism is unremarkable. Attachment facts are static. A street address changes once a decade, can be checked by anyone with a map, and costs nothing to publish — so it gets published, and it gets published well, with postcodes and suburb names.

Continuity facts are a different animal. A stated maintenance window is a promise that will be measured against reality next Tuesday. A redundancy claim invites a question about what happens when the second supervisor card also fails. An availability target implies staffing, an escalation path and somebody's job. A not-for-profit coordinator that publishes an undertaking has taken on an obligation; a coordinator that publishes an address has taken on nothing.

So the record is thick exactly where publication is free and thin exactly where publication is expensive. Anyone who has run a small exchange would predict this without being told.

What makes it matter is who bears the consequence. A fabric's failure does not stop at the contracting parties. Traffic from networks that peer through it, and from users who have never heard its name, reroutes or stalls according to decisions taken inside a membership they are not part of. That is not a scandal; it is what coordination infrastructure is. But it does establish the correct weight for record completeness: it matters most precisely where the affected population is wider than the consenting one.

Where a member's verification begins

At the edge of what the record publishes. That is not a shrug; it is the whole point. A member that depends on this fabric for domestic reachability is the party that carries the consequence, so verification of route-server reachability, of what happens when a session drops, of maintenance practice and of escalation falls to the member — by probing, by redundant sessions, by asking for the guide, and by reading the status page if it opens.

What an administrator can reasonably be asked for is bounded by the narrow coordination function: factual configuration data, kept current. Demanding a service-level agreement from a not-for-profit exchange fabric converts a coordination body into an assurance provider it never claimed to be, and the ask will be answered with silence or a document nobody staffs.

What would change the reading

A readable route-server address, with the fabric's own addressing. One sentence stating that the switching path is redundant, and what survives what. A dated availability or maintenance undertaking, or a status page that carries incident history rather than a green dot. The Technical Guide or Document Library opening and carrying the configuration layer in prose.

Any one of those changes the finding, and the finding should then be updated rather than defended. That is the honest arrangement: a dated negative with the conditions for its own retirement printed beside it.

Until one arrives, the position is narrow and defensible. The exchange publishes what is cheap to publish and verifiable by a site visit. What it does when a line fails is the part a member is actually buying.