IndexWhyAssetsCapabilitiesPlatformThesisNetworkResearchRegisterModelInvest
01 Semantic infrastructure for the superintelligence era

Somewhere
for machinesto stand

We build the canonical layer of the web. One address per concept, structured, cited and signed — so a model can resolve meaning instead of guessing at it. 77 properties live across 554 domains held outright, 451 of them single English words in .si.

Frontier models are trained on an open web that is filling with generated text carrying no source and no signature. The bottleneck on machine intelligence is stopping being compute and starting to be grounding.

0Endpoints owned
0Live properties
0Stated asset floor
0Developed asset value
02 / GROUNDING
02 Why this matters

Machines are about to inherit the world's knowledge from a web that can no longer prove anything.

Every frontier model is trained on the open web, and every agent answers from it. That corpus is now filling with generated text that carries no source, no author and no way to tell an original claim from its ten-thousandth copy. The bottleneck on machine intelligence is stopping being compute and starting to be grounding.

The failure mode
The fix is infrastructure

One concept. One address. One verifiable account.

The fix is not a better crawler or a bigger model. It is infrastructure: a layer of the web where each concept has exactly one canonical address, publishes structured meaning, cites its primary sources, and signs what it says — so that both a machine and a person can check it.

What it enables
Our position

We are not neutral about this and we do not pretend to be. We think the dictionary layer of the web is infrastructure, that it should be built to a verifiable standard, and that the window to assemble it coherently — rather than in ten thousand incompatible pieces — is now. We hold the addresses. The obligation that comes with them is to publish something worth citing.

Model collapse and the provenance gap are documented research problems; the claim that a canonical concept layer materially improves grounding is a design argument, not a measured result. We publish citation telemetry precisely so it can be tested rather than asserted.

Model collapse: Shumailov et al., Nature 2024 · AI crawl growth: Cloudflare Radar 2025 · Referral volume: Similarweb 2025 · Citation share: SearchSignal 2026. Full citations in the investment memorandum.

03 / LIVE NETWORK
03 Developed assets

77 properties live.
Every one of them, in one viewer.

No mockups. Pick any property and your browser fetches the real site, live. Each one sits at the exact lexical address of the concept it serves. Some hardened properties refuse embedding — the direct link is always there.

https:// live Open
Could not embed

The property did not respond inside the frame. Most hardened sites send headers that refuse embedding — expected, and why the direct link is always here.

Open in a new tab
fetching live document
Live document

Pick any property from the selector and its real site loads into this frame, fetched by your browser from the property itself. Nothing here is a rendering.

Open in a new tab
step through the network
Select a property
selection loads live
04 Capabilities

We build for the reader that is actually reading.

The web's principal consumer is no longer a person. Our engineering is aimed at the machines: verifiable provenance, structured meaning, and endpoints an agent can call rather than scrape. Six systems, shipping and in build.

How a property gets built

Every property runs the same six-stage pipeline, which is why the marginal cost of the next one approaches the cost of its content rather than the cost of a build.

05 The pattern

Every era of computing mints two letters into real estate.

Nothing about the asset changes. The market simply recognises what the namespace means — and reprices everything that ends in it.

A repricing needs four ingredients: a two-letter string matching an ascendant category of meaning; an open registration policy; a low entry price relative to that category's economic weight; and a catalyst that makes the match culturally legible. Anguilla had all four by 2017. Slovenia has the first three today — and the fourth is assembling itself in public.
06 Platform

One concept. One canonical address.

Conventional architecture hides topics behind paths. We invert it — each concept gets its own apex domain, so the concept's identity and its network address are the same string. Five layers, one templated stack, 554 endpoints.

retrieval gateway — illustrative

Illustrative session. Latency and document counts are examples, not measured production figures.

07 The network

554 domains. Owned outright. No claims.

Concentrated deliberately in the dictionary layer — roughly one in every 418 domains in the entire .si zone, and a far larger share of the English words inside it.

Already built
$1.82M

of stated value across 77 live properties

A quarter of the estate is already an operating business rather than dormant inventory. Development establishes bona fide use, earns and accumulates citations today, and means a buyer acquires a working property rather than a redirect.

Every name in the register is owned outright and unencumbered — no licence, no revenue share, no third-party claim.

Flagship assets
Why the machines change the arithmetic
Stated plainly: this argument is structural, not measured. We treat agentic demand as upside rather than as the base case — see risk 05.
Defensibility
08 Research

Eight public data points this thesis is priced against.

Our position rests on independent research into namespace economics — four decades of it, from the .com boom to Anguilla's .ai windfall. Fifty-three cited sources in the white paper.

Third-party public data, current as of July 2026. Not company results.

09 Register

Search the estate.

All 554 names, generated from the asset register. Blue means the property is developed and operating.

>
10 Business model

Four revenue lines. One estate.

A dictionary-word position has multiple exits. The word never expires, never rebrands, and is never devalued by a competitor's trademark.

Illustrative five-year build
Revenue EBITDA
Model shape
$4.72M2031 revenue
$9.36MCumulative, 5 years
2029EBITDA positive

Every line is unit-driven, so the assumptions are inspectable rather than asserted. Spend is concentrated in team, content and infrastructure — the lines that build the network.

An illustrative model, not a forecast and not realised revenue. It deliberately excludes portfolio appreciation, any single flagship sale, and any registry partnership.
Three scenarios, one bounded downside
All 554 names are already owned, so this is not an acquisition question. The reference class runs from a few hundred dollars to seven figures per name — and 77 developed properties earn while it resolves.
11 Investors

$3.50M at $18M post-money for 19.4%.

30 months of runway. The round converts a proven asset position into an operating network before the catalyst window closes.

Use of funds
What $18M post-money is priced against
Roadmap
The risk ledger

Each risk stated as its strongest version, with the actual mitigation rather than a reassurance.

The .com boom rewarded those who understood that businesses would need names. The next one belongs to those who understand that machine intelligence needs somewhere to stand.
network 77 live / 554 endpoints namespaces 6 manifests llms.txt · MCP · schema.org build 2026.08.10 --:--:-- UTC