A working draft for an open spatial web · July 2026

Hypertext → Hyperspace

Hypertext gave every document an address. Hyperspace gives every address a place. This is a research-grounded plan — protocol, toolkit, and working demo — for a metaverse that coexists with the entire internet instead of trying to replace it.

Spec CC0Code MITNo tokens · no registry · no owner
Draft 0.1 · prepared for David Clayton · ships with SPEC.md, world.schema.json, hyperspace.js, and a playable 4-world demo in one HTML file.

1The diagnosis: the metaverse died of architecture, not of boredom

Mid-2026 is a strange, clarifying moment to be writing this. In March, Meta announced the shutdown of Horizon Worlds' VR version — the flagship of an ~$80 billion bet — with the service ending June 15.1,2 Ready Player Me, the closest thing the industry had to a portable avatar layer, was acquired by Netflix in December and switched off six weeks later, stranding roughly 8,000 integrated apps.3 Spatial.io is discontinuing its creator platform this month; worlds that nobody downloads get deleted.4 Decentraland and Sandbox land — the "scarce digital real estate" thesis — sits 89–95% below peak.5 The tech press has written the obituary: "Well, there goes the metaverse."6

And yet. VRChat set an all-time concurrency record on New Year's Day 2026.7 Roblox reports 132 million daily users and paid creators over a billion dollars last year; Fortnite's creator islands crossed a billion in cumulative payouts.8,9 People demonstrably want shared virtual places. What died was not the desire — it was a specific architecture: the metaverse as a destination owned by a company.

Look at what every casualty had in common. Horizon Worlds lived in an app store, not at URLs — no links in, no links out, no search engines indexing it. Ready Player Me was a centralized service other businesses had to trust; one acquisition killed a whole interop layer. Land-based metaverses imported artificial scarcity into a medium whose superpower is abundance. None of them could be linked to, so none of them could benefit from the web's compounding network effects. Every one was a walled garden, and gardens die with their walls.10

This has happened before, and the postmortem is well documented. VRML — 1994's "VR equivalent of HTML" — shipped over 25 million bundled viewers and still died: authoring required 3D artists who didn't exist in web-dev numbers, plugins added friction, dial-up couldn't carry it, and there was no reason for site number one to publish a world when nobody was visiting worlds.11,12 Its co-creator Tony Parisi's own retrospective lesson: kill install friction, and let ordinary web developers author the content.12 Philip Rosedale, whose Second Life is still quietly running at a 20-year plateau, says the platforms keep chasing meetings and concerts when what retains people is "unusual creativity."13

The web is the only virtual world that ever reached everyone, and it won by being a layer over things that already existed — one file format, one link mechanism, viewers already on every machine. The metaverse will work the same way or it won't work.

2What history says actually makes a web layer stick

There is a repeatable playbook, visible across thirty years of formats that conquered the web without anyone's permission. robots.txt: one static text file, useful to the very first site that adopted it because the consumer (crawlers) already existed — it ran the web for 28 years before it was even formally standardized.14 Favicons: one file, zero configuration, instant visible payoff. Open Graph tags: four copy-pasted meta lines that one dominant consumer (Facebook) instantly rewarded with prettier links and more clicks — now present on roughly 60% of all webpages, the most successful optional layer ever shipped.15 Responsive design: no plugin, no new format, just existing CSS used progressively — then Google's 2015 "Mobilegeddon" ranking change made adoption economically mandatory overnight.16 HTTPS: stuck at ~58% until a free certificate authority with automated tooling removed both cost and toil; 81% within three years.17

And the anti-patterns: RSS died of fork politics and dependence on a single reader; PWAs stalled when one platform withheld the payoff; every "open metaverse" service (Hubs-as-a-service, Third Room, Webaverse, RPM) died when its patron company lost interest, while their self-hostable remnants survive.18,19

Distilled, the playbook that Hyperspace is built on: value at N=1 (one site benefits with zero network effects); one static file with no server changes; the viewer must already be installed — which in 2026 means the browser on five billion phones, not the headset on tens of millions of heads;20 the format must be crawlable so search and social can manufacture the incentive; progressive enhancement, never replacement; a tiny core vocabulary frozen early, with everything else as extensions; and no company, token, or registry anywhere load-bearing.

3The design: one file, four ideas

Hyperspace is deliberately small. A website joins the metaverse by serving one static JSON file and, optionally, one link tag — the favicon economics:

<link rel="world" href="/world.json">     (or /.well-known/world.json for the whole origin)

{ "hyperspace": "0.1", "name": "David's Studio" }

That two-member manifest is already valid: browsers render it as a generated plaza carrying the site's existing Open Graph card. From there it grows declaratively — a glTF scene if you have one, a generator if you don't, portals, spawn points, sky, panels, instancing policy — all data, no code, safe to fetch from strangers, cheap to index, impossible to weaponize. The full vocabulary fits on two pages of SPEC.md and freezes at 1.0.

The existing web (unchanged) HTML pages · URLs & links · HTTP · Open Graph · robots.txt · sitemaps · search engines The Hyperspace layer (one static file per site) world.json manifests · portals = links with a pose · cards = 2D boundary · branches = instances generators turn plain HTML into rooms · glTF scenes & VRM avatars where authored Any user agent, at its own level L1 a chat app shows the card · L2 a phone browser walks the world in a canvas L3 a headset browser goes immersive (WebXR) · L4 presence providers make it multiplayer

Four ideas carry the whole thing.

Portals are links. A portal is a hyperlink with a position — walking through it is the same act as clicking it, with the same history, back button, and bookmarks. Trust chrome is mandatory: every portal displays its destination origin, and no world may move your camera without a traversal gesture, which closes the spatial phishing class at the spec level. Deep links address positions inside worlds through URL fragments (#pos=4,0,-2&yaw=90), a lineage borrowed with credit from the XR Fragments draft.21

Generators make the whole web eligible. VRML proved you can't staff a metaverse with 3D artists. So the spec's most important scene format isn't a scene at all: a generator is a named, deterministic procedure that builds a world from the page's own semantics. hyperspace:reading-room parses any article's HTML — headings become wall panels in reading order, figures get framed, links become labeled doors. Semantic HTML has been a floor plan all along; the demo's Reading Room is built live from a real article by exactly this pipeline. A site with no 3D budget gets a dignified world for free; a site with ambition ships glTF.

Cards keep the metaverse indexable. Every world must project to 2D — name, description, image — so search engines, chat apps, and flat browsers see into the spatial web without rendering it. This is the Open Graph flywheel aimed at worlds: the day one significant consumer displays cards and counts portal links, every site on earth has a reason to add the file. It also means the metaverse gets PageRank-able topology — portals form a link graph — and anyone can crawl an Atlas; the spec defines no registry, and an index you must pay to join is simply not the Atlas.

Branches abolish both crowding and scarcity. A URL names a world, not a room. Runtimes spawn as many concurrent branches as demand requires; travelers sharing an entanglement token (an opaque secret in #branch=…) always resolve to the same one — sharing a link is the invitation. Capacity limits stop meaning "the metaverse is full" and start meaning "you got a fresh branch." There is no land, no plot map, no artificial scarcity to speculate on; address space expands like the universe. That is a design refusal written directly against the land-crash evidence.5

4The physics, used honestly

You asked whether frontier physics could inform this, and it turned out to be more than decoration — several structures from physics are exact solutions to protocol problems, and borrowing their names makes the spec teachable:

Hyperspace termPhysics sourceWhat it solves
Atlas & chartsDifferential geometry — a manifold has no global coordinates, only overlapping local chartsNo world-coordinate system, no land registry; worlds are local charts stitched by portals. The "one global map" instinct is what created land speculation.
PortalTraversable wormhole — two mouths, zero distanceAdjacency without geometry: any place can neighbor any place, which is just what a hyperlink always was.
BranchEverett's many-worlds — identical content, decohered historiesInstancing semantics: same world, independent state, unlimited copies.
Entanglement tokenCorrelated states resolve identically when measuredParty invites: same token ⇒ same branch, wherever and whenever you enter.
CardHolographic principle — a volume's information encoded on its boundaryEvery 3D world fully summarized on its 2D surface, keeping the metaverse searchable from flatland.
Causal patchLight cones bound what an observer can be affected byInterest management: full-fidelity sync only near you; the demo's Observatory makes it a toggle you can walk on.
Observation collapseUnobserved systems remain in superpositionLazy loading with dignity: distant content is impostor haze until approached.
ExpansionThe universe grows without center or edgeThe anti-scarcity clause: there is always more space, so space is never for sale.

5Why 2026 is finally the right year

Every building block this needs already shipped, most of it in the last eighteen months. WebXR reached Candidate Recommendation Draft in June;22 WebGPU is now in all major browser engines;23 glTF is the ISO-standard delivery format with ratified compression, near-final interactivity graphs, and Gaussian splats at release-candidate;24 VRM avatars are being standardized through Khronos itself.25 And crucially, the standards world is independently converging on this exact architecture: the Metaverse Standards Forum's Web of Worlds whitepaper (June 2, 2026) argues for worlds as URI-addressable web endpoints, and its new open-source "Sneeze" browser engine follows two weeks later.26,27 Hyperspace is positioned as the minimal publisher-side on-ramp to that same future — the two-line file a WordPress plugin can emit — and the spec commits to harmonizing with the MSF work and contributing the draft, CC0, to the open venues.

Where people are, July 2026WebXR VRHandsNotes
Meta Quest BrowseryesyesFullest implementation; passthrough AR, anchors, layers28
Apple visionOS Safariyesgaze+pinchOn by default since visionOS 2; <model> element shipped in visionOS 2629,30
Chrome on Android XR (Galaxy XR)yesyesLaunched Oct 2025 with strong WebXR support31
Wolvic / PICO browsersyesvariesOpen-source Wolvic actively maintained32
Desktop Chrome / Edgeyes—Via OpenXR; flat L2 mode everywhere regardless
Firefox · Safari (Mac/iOS)no—This is why L2 flat mode is mandatory: phones are the metaverse's front door

The hardware honesty matters too: headset shipments declined 14–16% year-over-year, Vision Pro sold roughly 45,000 units in Q4, and Meta is pivoting to smart glasses.20,33 A protocol that requires a headset is a protocol that requires a market that doesn't exist yet. Hyperspace is phone-first with headsets as the upgrade — which is exactly how the web treated broadband, color screens, and GPUs.

6What got built (and verified) today

The Hyperspace Atrium: a night court with an obelisk, four glowing portal rings, ghost travelers, and a HUD address chip
The Atrium — hub world of the demo. Four wormhole portals with mandatory origin labels, pseudonymous ghost travelers, and the address chip (trust UI) up top. This is one self-contained HTML file running the open-source viewer.

The starter kit in your hands contains a complete vertical slice of the protocol, all of it exercised by an automated headless test suite (14 checks, all passing, zero console errors):

SPEC.md — the Working Draft: discovery, manifest vocabulary, portals and trust rules, branches, presence, security/privacy/accessibility requirements, conformance ladder, prior-art mapping, and the physics lexicon. world.schema.json — machine-validatable JSON Schema (the kit's example manifests validate against it). hyperspace.js — the MIT reference viewer: ~1,200 lines of dependency-light three.js that any site can drop in; it discovers <link rel="world">, renders manifests and glTF scenes, runs the three core generators, handles portals with fade traversal and history, speaks touch, mouse/keyboard, and WebXR with thumbstick locomotion and snap turn, and exposes the whole thing as a small API.

The demo — hyperspace-demo.html, four linked worlds in one 800 KB file with no external dependencies: the Atrium (hub, ghosts, a live embedded flat website where the environment permits); the Flatland Gallery (how 2D pages appear in space, plus the Atlas table and an honest wall about browser security); the Causal Patch Observatory (walk-on branch pads that hop many-worlds instances, the causal-patch LOD toggle, a holographic card exhibit, and an orrery); and the Reading Room, generated at runtime from a real HTML article by the same parser any spatial browser would use. One of the portals fetches its destination world over an actual URL at traversal time — the full network pipeline, demonstrated inside a single file.

The Causal Patch Observatory: orrery with orbit rings, two portal rings, plaques, and branch pads under a starfield
The Observatory: many-worlds branch pads (right), wormhole portals with origin chips, and physics plaques that double as protocol documentation.
The Reading Room: a warm circular room with a title frieze, section panels on the walls and reading lamps
The Reading Room — built live from an HTML article by the hyperspace:reading-room generator. Headings become panels, links become doors. Any blog post can be this, with zero 3D authoring.

Trying it

On a phone or laptop, just open the file: left half of the screen (or WASD) walks, dragging looks, glowing rings traverse. On a Quest or any WebXR headset, the file needs to be served over https — put it on GitHub Pages, Neocities, or any static host, open that URL in the headset's browser, and press Enter VR. (WebXR requires a secure origin; that's the only reason a plain file won't do.) The ◈ Worlds drawer is the pocket Atlas — it can also load any world.json URL you paste.

7The adoption ladder

Each phase is independently useful, none requires permission, and every one is copied from a move that already worked once. Phase 0, the polyfill (this kit): any site adds the file and a script tag and is walkable in every 2026 browser — value at N=1, viewer pre-installed on five billion devices. Phase 1, the toolmakers: plugins for WordPress, Squarespace, itch.io, and static-site generators emit manifests automatically — favicon economics, millions of worlds from a handful of integrations. Phase 2, the Atlas: open crawlers index cards and portal graphs; the first search or social product to render a "Worlds" tab manufactures the incentive for everyone (the Mobilegeddon move). Phase 3, the browsers: VR browsers add native rel="world" handling — an "Enter" button in the URL bar — and browser-owned panels complete the 2D-in-3D contract; flat browsers ship a spatial reader mode. Phase 4, together: presence bindings (WebRTC mesh, Matrix, WebTransport) standardize, and the web is multiplayer where it wants to be and solitary where it doesn't.

There is no moat, and that is the moat. Anyone can implement it, fork it, index it, or sell services on top of it — which is precisely why it can win where every owned metaverse lost.

8Objections, answered

Isn't this just VRML again? VRML demanded a parallel content universe from authors who didn't exist, viewed through plugins nobody had. Hyperspace generates worlds from content that already exists, viewed in browsers everyone already has. The failure mode was the delivery mechanics, not the dream.11

Why glTF and not USD? Delivery versus authoring. OpenUSD (Core Spec 1.0, Dec 2025) is winning interchange and simulation; glTF is the ISO royalty-free runtime format every web engine loads today.34,24 Author in USD, deliver glTF — the spec says so explicitly.

Why JSON instead of new HTML tags? A separate file is discoverable by crawlers without rendering, cacheable, servable by any static host, attachable to non-HTML resources through a header, and — critically — adoptable by pasting one file, which is the single strongest predictor of web-layer success (§2).

Why no blockchain? Because scarcity is the failure mode, not the feature; the evidence is nine-figure losses on virtual land.5 Identity and payments remain pluggable extensions for those who want them; nothing in the core requires them.

Doesn't MSF's Web of Worlds make this redundant? It makes it timely. WoW is the full-stack architecture conversation among vendors; Hyperspace is the publisher-side minimal profile — the on-ramp a solo developer can adopt this afternoon. The draft commits to harmonizing vocabulary and contributing to those venues rather than competing with them.26

What's genuinely hard and unsolved? Two things, named honestly in the spec. Live cross-origin webpages as in-world textures are impossible from page JavaScript by security design — only browser vendors can complete that (it's Phase 3's ask, and the demo shows the honest fallback). And presence at scale needs real servers; the spec keeps that layer pluggable so no single provider can ever become the next Ready Player Me.3

9Governance, or the art of refusing things

The spec text is CC0 — public domain, forkable without asking. The reference code is MIT. The draft is written to be contributed to the W3C Immersive Web Community Group and the MSF 3D Web Interoperability WG. And Appendix C of the spec is a list of refusals: no coin, no chain, no land sale, no registry, no required account, no flagship app, no owner. Every item on that list was load-bearing in some prior failure. The protocol's only asset is that nobody — including its authors — can capture it.

"Hypertext gave us pages. Hyperspace gives us places. The rest is just world.json."

10Sources

  1. CNBC — Meta is shutting down VR Horizon Worlds (Mar 18, 2026): cnbc.com/2026/03/18/meta-horizon-worlds-metaverse-vr.html
  2. CNBC — Reality Labs Q4-2025 loss $6.02B (Jan 28, 2026), cumulative ≈$80B: cnbc.com/2026/01/28/metas-reality-labs-posts-6point02-billion-loss-in-fourth-quarter.html
  3. R. Schultz — Netflix acquires Ready Player Me; service shut down Jan 31, 2026: ryanschultz.com/2025/12/22/…; genies.com/blog/ready-player-me-discontinued-alternatives
  4. UploadVR — Spatial is discontinuing its creator platform in July 2026: uploadvr.com/spatial-is-discontinuing-its-creator-platform-in-july/
  5. CryptoSlate — Metaverse land values capitulate; Sandbox −95%, Decentraland −89% (Mar 19, 2026): cryptoslate.com/rip-metaverse-land-values-capitulate…
  6. TechCrunch — "Well, there goes the metaverse" (Jan 19, 2026): techcrunch.com/2026/01/19/well-there-goes-the-metaverse/
  7. UploadVR — VRChat all-time concurrency record ~149K (Jan 2026): uploadvr.com/vrchat-nye-user-record/
  8. Roblox Q1-2026 shareholder letter — 132M DAU; 2025 creator payouts >$1B: sqmagazine.co.uk/roblox-statistics/; about.roblox.com/newsroom (Apr 2026)
  9. Tubefilter — Epic passes $1B cumulative UEFN island payouts (Jun 17, 2026): tubefilter.com/2026/06/17/epic-games-unreal-editor-for-fortnite-creator-payouts/
  10. J. Radoff — Metaverse Interoperability: Challenges (2022); R. Zuckerberg on walled gardens (CNBC, 2022): medium.com/building-the-metaverse/…; cnbc.com/2022/08/18/…
  11. How-To Geek — Remembering VRML: The Metaverse of 1995: howtogeek.com/778554/; ClickZ — The Rise and Fall of VRML (2000)
  12. UploadVR — Tony Parisi on co-creating VRML (retrospective lessons): uploadvr.com/flashback-tony-parisi…
  13. Skarred Ghost — Philip Rosedale interview (Jun 2025): skarredghost.com/2025/06/19/philip-rosedale-quick-interview/
  14. RFC 9309 (2022), robots.txt history: datatracker.ietf.org/doc/html/rfc9309
  15. Web Almanac 2024, structured data — OG on ~59–64% of pages: almanac.httparchive.org/en/2024/structured-data
  16. A List Apart — Responsive Web Design (2010); "Mobilegeddon" ranking update (Apr 21, 2015)
  17. Let's Encrypt — One Billion Certificates (Feb 2020), HTTPS 58%→81%: letsencrypt.org/2020/02/27/one-billion-certs
  18. Two-Bit History — The Rise and Demise of RSS (2018): twobithistory.org/2018/09/16/the-rise-and-demise-of-rss.html
  19. Mozilla — End of support for Hubs (May 2024); hubsfoundation.org; Third Room paused (Matrix, Jul 2023); JanusWeb v1.7.2 (May 2026): github.com/jbaicoianu/janusweb
  20. Forbes — Quest shipments −16% YoY, IDC (Jan 5, 2026): forbes.com/sites/andrewwilliams/2026/01/05/…
  21. XR Fragments draft (L. van Kammen, 2023; archived 2025): xrfragment.org/doc/RFC_XR_Fragments.html
  22. W3C — WebXR Device API, Candidate Recommendation Draft (Jun 9, 2026): w3.org/TR/webxr/
  23. web.dev — WebGPU now supported in all major browser engines (Nov 2025): web.dev/blog/webgpu-supported-major-browsers
  24. Khronos — glTF interactivity extensions approach finalization (May 2025); KHR_gaussian_splatting RC: khronos.org/blog/…; github.com/KhronosGroup/glTF
  25. Khronos × VRM Consortium collaboration (Oct 24, 2024): khronos.org/news/press/…vrm-3d-avatar-file-format
  26. Metaverse Standards Forum — Web of Worlds whitepaper (Jun 2, 2026): metaverse-standards.org/news/blog/announcing-the-web-of-worlds-whitepaper…
  27. MSF — "Sneeze," first open metaverse browser engine, Apache-2.0 (Jun 15, 2026): metaverse-standards.org/news/press-releases/…sneeze…
  28. Meta Quest Browser release notes (v146, 2026): developers.meta.com/horizon/documentation/web/browser-release-notes/
  29. UploadVR — visionOS 2 enables WebXR by default (2024): uploadvr.com/visionos-2-apple-vision-pro-webxr/
  30. WebKit — The HTML model element in Apple Vision Pro (visionOS 26, 2025): webkit.org/blog/17118/
  31. Android Developers — WebXR on Android XR: developer.android.com/develop/xr/web; Google — Samsung Galaxy XR launch (Oct 2025)
  32. Igalia — Wolvic releases: github.com/Igalia/wolvic/releases
  33. 9to5Mac — Vision Pro ≈45K units in Q4 2025, IDC (Jan 2, 2026): 9to5mac.com/2026/01/02/…
  34. AOUSD — OpenUSD Core Specification 1.0 (Dec 17, 2025): aousd.org/news/core-spec-announcement/

Prepared July 17, 2026, in one working session: three parallel research sweeps (standards, market, adoption history), then spec → schema → library → demo → automated verification (14/14 checks, zero console errors). Everything in the kit is yours to publish, fork, or submit — the licenses are the whole permission system.

Hyperspace is a working title; "the protocol" is the two files: world.json and the link that points at it.