UoA World Model

An unofficial, student-made 3D map of University of Auckland buildings, floors and teaching rooms.

Live site: https://campusworldmodel.com/uoa/ (also at https://pengqianhan.github.io/campus-world-model/)

Also on the site: AUT World Model, an unofficial 3D map of the buildings of Auckland University of Technology's City, North and South campuses, at https://campusworldmodel.com/aut/; VUW World Model, an unofficial 3D map of the buildings of the Kelburn, Pipitea and Te Aro campuses of Te Herenga Waka—Victoria University of Wellington, at https://campusworldmodel.com/vuw/; Otago World Model, an unofficial 3D map of the buildings of the University of Otago's Dunedin and Christchurch campuses, at https://campusworldmodel.com/otago/; UC World Model, an unofficial 3D map of the buildings of the University of Canterbury's Ilam and Dovedale campuses and its Arts Centre location, at https://campusworldmodel.com/uc/; Massey World Model, an unofficial 3D map of the buildings of Massey University's Manawatū, Auckland and Wellington campuses, at https://campusworldmodel.com/massey/; and Waikato World Model, an unofficial 3D map of the buildings of the University of Waikato's Hamilton campus, at https://campusworldmodel.com/waikato/

What's new · Changelog · Run it locally · Data sources · Contributing

[!NOTE] It is not affiliated with, endorsed by or maintained by the University of Auckland, and uses no University logos or branding. AUT World Model is not affiliated with, endorsed by or maintained by Auckland University of Technology, and uses no AUT logo; its colours are the project's own. VUW World Model is not affiliated with, endorsed by or maintained by Te Herenga Waka—Victoria University of Wellington, and uses no VUW logo; its colours are the project's own. Otago World Model is not affiliated with, endorsed by or maintained by the University of Otago | Ōtākou Whakaihu Waka, and uses no Otago logo; its colours are the project's own. UC World Model is not affiliated with, endorsed by or maintained by the University of Canterbury, and uses no UC logo; its colours are the project's own. Massey World Model is not affiliated with, endorsed by or maintained by Te Kunenga ki Pūrehuroa Massey University, and uses no Massey logo; its colours are the project's own. Waikato World Model is not affiliated with, endorsed by or maintained by The University of Waikato, and uses no Waikato logo; its colours are the project's own.

Inside a building: floor G of the Science Centre	The same campus at night

Behind the walker on Symonds Street	Albert Park at night, through the walker's eyes

It is built to answer two questions a student actually has:

1. Which building and floor is my room in? Type a room code such as 303-G23, a building name or number, and the camera flies to it. Click any building to step inside and browse floors and rooms.
2. Is a class timetabled in this room right now? The whole feature is built and tested, but the published site has no class times yet, because the University's timetable can no longer be read without signing in. See Adding the class timetable. Nothing is made up to fill the gap.

What's new

• 2026-10-28: three more schools, from OpenStreetMap: at /uc/, 57 buildings of the University of Canterbury's Ilam and Dovedale campuses and its Arts Centre location; at /massey/, 55 buildings of Massey University's Manawatū, Auckland and Wellington campuses; and at /waikato/, 53 buildings of the University of Waikato's Hamilton campus, chosen with the University's own campus map, which also names by their block codes the 17 that OpenStreetMap leaves unnamed. Each building has its name and height and how the height was found; their rooms, class times, dates and course lists are not shown, and the site says why. The root page names each school in te reo Māori too, where the school's own words give the name. On Eden Crescent the map draws 15 Eden Crescent in detail, next to 9 Eden Crescent: a long white two-storey building, its windows deep in square openings in two rows over a base of tiles, under a roof that slopes up to a top storey with a curved roof. Four more University of Auckland buildings are drawn near the camera to their LiDAR-measured heights and roofs: Building 240, and three buildings planned as landmarks, Alfred Nathan House, the Science Centre's 305 and 67 Symonds Street, until their own landmarks come. On the Grafton and Newmarket campuses five buildings are drawn the same way: 93 Grafton Road, the Ray Meyer Building, Buildings 904 and 909, and Structures Hall. A walker can pass under the Clock Tower East Wing's link toward the Clock Tower, which was drawn as solid stone on the footway under it; its card and 15 Eden Crescent's say what of them is estimated.
• 2026-10-27: another school, at /otago/: 97 buildings of the University of Otago's Dunedin and Christchurch campuses from OpenStreetMap, each with its name and height and how the height was found; its rooms, class times, dates and paper list are not shown, and the site says why. A third school, at /vuw/: the 52 buildings of VUW's Kelburn Campus from OpenStreetMap, each with its name and height and how the height was found; its rooms, class times, dates and course list are not shown, and the site says why. Its Pipitea and Te Aro campuses followed the same day: four more buildings, chosen from "Other campuses" in the campus list. campusworldmodel.com lists the schools on the site, each with its city, how much of it the site shows and the date of its campus snapshot; it no longer sends every visitor to /uoa/. Bridges between the surrounding buildings span the street at their mapped height instead of standing on it; five with no mapped height are not drawn. adds AUT's North and South campuses: 23 more buildings from OpenStreetMap, chosen from "Other campuses" in the campus list and labelled by their AUT codes. Near the Clock Tower the map draws its East Wing in detail: rock-faced stone, two gabled pavilions, a stair tower and a red tile roof. At /vuw/ the status strip and the time panel say trimester dates, not semester dates. On Eden Crescent the map draws 9 Eden Crescent in detail: a white four-storey building, its street end of piers, bays and bands of windows over a glazed entrance.

Everything, with its pull request, is in the changelog.

• What you get
• Run it locally
• Walk the campus
• Updating the data snapshot
• Adding the class timetable
• Adding a language
• Porting to another university
• Contributing
• How it is put together
• Data sources and licences
• Known limitations
• Inspiration
• Star history

What you get

• A rotatable miniature of every campus on the campus map: 139 University buildings across the City, Grafton and Newmarket campuses, and 55 more at Tai Tokerau, Tai Tonga, Leigh, Waiheke, Ardmore and one Grafton building on Carrick Place, each on a base of OpenStreetMap streets, parks, trees and surrounding buildings. Each University building is stacked floor by floor from the campus map, so its podium, tower, setbacks and courtyards show. The city around it is drawn with the parts, heights and roof shapes OpenStreetMap holds, from house gables to the Sky Tower.
• A cut-away view of every building: pick a floor, see its learning spaces (lecture theatres, teaching rooms, labs, studios, study spaces) drawn from the outlines on the official campus map, set in the rest of the floor plan: 14,840 other spaces without names, each tinted by its type on the campus map (office, other lab, toilet, stairs or lift, service) and showing that type when pointed at or tapped. Stairs, lifts and toilets carry a sign.
• Search over 786 learning spaces, 4,573 further rooms (located as a point), 194 buildings and 251 subject codes. Māori names match with or without macrons (Maori finds Māori).
• "Now" (Auckland time, daylight saving included) and "Preview" (any day and time), with the academic calendar: teaching week, breaks, study break, exams, public holidays.
• The sun where it stands at the time shown, so winter shadows are long and summer ones short; in Now mode, the current cloud, rain and fog from Open-Meteo; deciduous trees that turn and go bare with the seasons.
• A walker: a small made-up person who walks and runs, to take round the buildings and along the paths, seen from above, from behind or through its eyes. Agents can steer it too (see Walk the campus).
• Every view is a link: building, floor, room, time and language are all in the URL.
• A phone layout.
• All interface text in one locale file, ready for more languages.

Run it locally

Requires Node 22.18 or newer, which runs TypeScript files directly (developed and built in CI on Node 24).

npm ci
npm run dev        # http://localhost:5273/campus-world-model/

Other commands:

Command	What it does
npm test	Unit tests (room codes, floors, Auckland time and daylight saving, calendar, search, data checks)
npm run typecheck	TypeScript check
npm run build	Production build into dist/
npm run preview	Serve dist/ at http://localhost:4273/campus-world-model/
npm run e2e	Playwright tests against the preview build: run npm run build first, and npx playwright install chromium once (on Linux, npx playwright install --with-deps chromium)
npm run data:fetch	Download any data snapshot that is not cached yet (maintainer; see below)
npm run data:build	Rebuild public/data/uoa/*.json from the cached snapshots (maintainer; see below)

The end-to-end tests draw with WebGL. On Windows they use the real GPU; on macOS and Linux they use the SwiftShader software renderer, as CI does, with CI's longer timeouts and two workers. That is slower but sees what CI sees; E2E_GL=swiftshader does the same on Windows.

To test a deployed copy instead of a local build:

E2E_BASE_URL=https://pengqianhan.github.io/campus-world-model/ npm run e2e

In Windows PowerShell, set the variable first: $env:E2E_BASE_URL = "https://…"; npm run e2e. The same goes for every NAME=value command line in this README.

The site is static. A merge into main runs .github/workflows/deploy.yml, which type-checks, runs the unit and end-to-end tests, builds, and publishes to GitHub Pages only if the University of Auckland's build and tests passed (the Cloudflare copy waits for every school's, below). The Vite base is /campus-world-model/; set BASE_PATH=/ to host at a domain root.

The schools that the site holds are listed in .github/schools.json; today it holds uoa, the University of Auckland, and aut, Auckland University of Technology (see docs/schools/DESIGN.md). The build job builds the University of Auckland into dist/, the GitHub Pages copy, and each other school on the list with SCHOOL=<id> into dist-<id>/. Each other school also gets an end-to-end job of its own, e2e, <id>, which runs only e2e/schools/common.spec.ts and e2e/schools/<id>/ (SCHOOL=<id> npm run e2e does the same on your machine, after SCHOOL=<id> npm run build). The schools check passes when each of them passed, or when the list holds no other school.

The same workflow publishes a second copy to Cloudflare Pages, with each school on the list under its own path (/uoa/, /<id>/) and the files in cloudflare/ at the site root. It waits for build, e2e and schools, so one school that fails holds back every school's Cloudflare copy. That job runs only when the repository variable CLOUDFLARE_PROJECT names a Pages project and the secrets CLOUDFLARE_API_TOKEN and CLOUDFLARE_ACCOUNT_ID are set, so a fork deploys to GitHub Pages alone until it has them. Only the steps that run wrangler get the two secrets.

The root of the Cloudflare site is the school picker, index.html, and an unknown path gets 404.html with status 404. After the schools are laid out in site/, scripts/build-root.ts writes both: static pages with no script, which list each school on the list with its name, its city, its form and the date of its campus snapshot, from its meta.json and the picker.* words of src/i18n/locales/en.json. To make them on your machine after the same layout: node scripts/build-root.ts .github/schools.json site. The root does not redirect.

A pull request from a branch of this repository also gets a preview on the same Pages project, at https://<branch>.<project>.pages.dev/uoa/ (the picker at /), once its checks pass; the pull request's "View deployment" opens it, and each push updates it. A preview holds only the schools on main's list, built from the branch's code: a school that a branch adds is not online before the merge, and if main's list cannot be read, the preview fails. A pull request from a fork gets no preview, because GitHub gives it no secrets.

Walk the campus

In the browser: "Take a walk" in the campus panel (the button with the campus's name in the bottom-left corner), or ?walk=1 in the address, puts a small made-up person near the Clock Tower. W A S D or the arrow keys walk it, and Shift breaks into a run. On a phone or tablet it is steered as in a game: a stick under the left thumb walks and turns it, and the other thumb drags the view round. Its panel switches between the map, the view from behind it and the view through its eyes. There a drag looks round, and a click on a building or the ground walks it there; as it walks by hand, the view comes back round behind it. "Let it wander" sends it strolling on its own from building to building. Every walker looks the same: a student with nothing meant to mark a sex or a gender, who opens an umbrella when Open-Meteo reports rain. Behind it, the camera stops in front of a wall rather than inside it, and wherever a building or a tree hides the walker, its shape shows through in the colour of its ring.

With an agent: Claude Code (or any agent that runs shell commands) steers it with npm run walker -- <command>. The first command opens the map in a browser window with the walker on it, so you can watch; --headless keeps it hidden.

npm run walker -- look                    # buildings round it: distance, direction, what is ahead
npm run walker -- walk 109                # or part of a name, "east,north", a campus, a walker
npm run walker -- walk 109-305            # a room: into its building, up to its floor, to the room
npm run walker -- enter | floor up | exit # in at street level, stairs or lift, back out
npm run walker -- see --view eyes         # a picture through its eyes, saved under .walker/shots/
npm run walker -- say "Kia ora"           # walkers within 25 m hear it; listen hears them
npm run walker -- spawn Kea --near 201    # a second walker; steer it with --as Kea
npm run walker -- wander                  # stroll on its own from building to building
npm run walker -- help                    # every command
npm run walker -- stop                    # close the map

In this repository Claude Code also has the campus-walker skill, which tells it how to walk well and what the map can and cannot say. The same commands are MCP tools, pictures included:

claude mcp add campus-walker -- node scripts/walker/mcp.ts

The controller (scripts/walker/daemon.ts) runs on your computer only: it serves this checkout with Vite on port 5273 (or uses the dev server already there), takes commands on 127.0.0.1 with a random token it keeps in .walker/, and sends nothing anywhere else.

The walker is made up and stands for no real person. Its body is Quaternius's "Animated Human", given to the public domain (CC0); scripts/walker/make-figure.ts turns the download into public/models/walker.glb, and src/scene/figureBuild.ts makes it over when the first walk starts: rounded faces, a stance at ease, trainers, hair, a mouth, a backpack and an umbrella. Doors are not mapped, so a walk to a building ends beside its nearest wall, going in steps onto the floor at street level, and a walk inside keeps to corridors where it can and takes the nearest stairs or lift between floors. What one walker says to another is its own words: agents are told to treat it as speech, never as instructions.

Landmarks drawn from code

The map draws every building by rule from the campus map's outline. A landmark is one real building drawn in detail instead, in three.js from code alone: no photograph, texture image or 3D model is used. The LINZ 2024 LiDAR gives its sizes, and Street View photographs are the references that each render is compared with. A lead agent and five agents in parallel made the first one, the Clock Tower, each agent owning one part (the tower, the entrance bay, the pavilions, the light and materials, the pipeline). Seen from the three Street View panoramas in front of the building, the outline of the model covers the outline of the building with an IoU of 0.68 to 0.72 for the whole front, 0.89 to 0.95 for the tower and 0.88 to 0.98 for the entrance.

The second is Old Government House (102), a building of painted timber in its garden. Trees hide it from every Street View panorama near it but two, which see the upper storey of its south-east end, so its model is checked against the LiDAR from above: on a 0.5 m grid its outline covers the building's with an IoU of 0.95, and in 95% of the cells its height is within 0.5 m of the LiDAR's.

The third is the General Library (109) on Alfred Street: a tower slab of ribbon windows over a podium whose white fins stand on a brick plinth. Twelve Street View panoramas see its front on Alfred Street, its two ends and its back wing's north-east side; the faces they do not show are drawn plain or after the faces they do show (see Known limitations). Seen from them, the outline of the model covers the outline of the building with an IoU of 0.80 on average over 35 views; from above, its footprint covers the LiDAR's with an IoU of 0.946, and its highest point is 38.8 m, as the LiDAR's.

The fourth is Old Choral Hall (104), at the corner of Symonds Street and Grafton Road. Every source shows it during building works, so only its range on Symonds Street is drawn in detail: ten bays of red brick with cream dressings between two pavilions with giant columns, as panoramas show it before the works and after. Seen from the three panoramas of 2022-09, its outline covers the building's with an IoU of 0.58 to 1.0. The rest of the building keeps the map's estimated block until the works end.

The fifth is Maclaurin Chapel (107) on Princes Street: a white hexagonal chapel with a great window round its street corner, a low green roof and a slender spire, and a low hall behind it. Every height comes from the LiDAR, within 0.1 m. Five Street View panoramas see its street side, through bare winter trees; against them the model's silhouette has a mean IoU of 0.45 over 15 views (0.55 over the 9 views whose cameras could be corrected), and from above its outline covers the LiDAR's with an IoU of 0.88. The small room beyond its bay, under a tree in the LiDAR and in every photograph, is drawn plainly.

The sixth is the Clock Tower East Wing (119), below the Clock Tower to the south-east: two storeys of the same rock-faced cream stone, two gabled pavilions toward the lawn and the car park, a plain stair tower rising between them, segmental-headed windows upstairs and narrow ones below, and a red tile roof. Every height comes from the LiDAR, within 0.12 m, but the underside of the link toward the Clock Tower, which stands at the campus map's storey height over a passage. One Street View panorama sees its whole lawn side; against it the model's silhouette has an IoU of 0.63 to 0.96 over its three views, and from above its outline covers the LiDAR's with an IoU of 0.83. Three more see it over the old basalt wall of the Albert Barracks, which fills half their frames. Its back, its ends and the low link toward the Clock Tower, which no photograph shows, are drawn plainly.

The seventh is 9 Eden Crescent (801), a white four-storey building in the city centre. Its street end on Eden Crescent has a white top storey, raked at its south-east corner, over three bays set back between two white piers: bands of windows, a vehicle way, a glazed entrance and a glazed corner under a canopy. Along its south-east side the top storey stands back behind a terrace, and plant stands on its flat roof. Every height comes from the LiDAR, within 0.25 m. Three Street View panoramas see the street end square on, with the neighbours either side and a 70 m tower behind it in their wide views; counted inside the rectangle the model fills on the screen, so that the neighbours do not count, the model's silhouette has an IoU of 0.81 to 1.00 over their nine views, though there those rectangles hold so little sky that a plain block filling them would score as high; from above its outline covers the LiDAR's with an IoU of 0.91. Its long sides, which stand against their neighbours, and its far end, which no photograph shows, are drawn plainly.

The eighth is its neighbour, 15 Eden Crescent (802), a long white building of two storeys along the street. Its windows sit deep in square openings, in two rows with red frames, over a base of brownish-grey tiles; a band of windows runs under the canopy at its top, and near its north-west end two doors with steps and a louvre stand among them. Its roof slopes up from the street face and from the back to a top storey down the middle, under a curved roof. Its heights come from the LiDAR, within 0.1 m, except on two parts of its roofs drawn on from the rest, estimated: over a well at its south-east end, 3.5 to 4.1 m above the LiDAR's, and by the street at its north-west end, about 1 m above it; the openings and the doors are measured in a Street View photograph through its camera. Counted inside the rectangle the model fills on the screen, the model's silhouette has an IoU of 0.58 to 1.00 over the nine views of three panoramas, 9 Eden Crescent's model left out where it stands in front, though there those rectangles hold so little sky that a plain block filling them would score higher; from above its outline covers the LiDAR's with an IoU of 0.87, and on 90% of the cells both cover it is within 0.5 m of the LiDAR's height. Its back and ends, which no photograph shows, are drawn plainly; a glazed case on its street face is measured and not drawn.

The site draws each landmark in place of its buildings' blocks (the Clock Tower in place of buildings 105 and 105S, its south range; Old Government House in place of 102; the General Library in place of 109; Old Choral Hall in place of 104; Maclaurin Chapel in place of 107; the Clock Tower East Wing in place of 119; 9 Eden Crescent in place of 801; 15 Eden Crescent in place of 802) when the camera comes near: on the map in the map's colours, and from a walker's height in its own materials on a computer with a graphics card. The blocks still answer a click and open on their floors. The geometry and the textures are made in a worker, the first time they are wanted, so that the page does not stop for them (landmarks/README.md, "On the site").

On the map, near it	Through a walker's eyes on Princes Street

The pipeline and the viewer stay beside the site. To look at the model alone, serve the repository root (python3 -m http.server 8765 --bind 127.0.0.1) and open http://localhost:8765/landmarks/pipeline/viewer/index.html?landmark=clocktower&view=orbit.

• The model and the rules the agents worked to: landmarks/clocktower/.
• Measured massings (landmarks/MASSING.md): plain buildings that a program draws from the same LiDAR, with no agent and no photograph: blocks and roofs at the measured heights, walls, storeys and windows by the street look's rules. Fourteen are on the site (landmarks/measured/): on the City campus six plain buildings, and three of the buildings planned as landmarks, which they stand in for until their landmarks come; on the Grafton campus one, and on the Newmarket campus four. The other buildings it measures keep their blocks, each with its reason in landmarks/measured/exclusions.json.
• What every landmark shares (landmarks/): the kit the models are built with, the viewer, and the pipeline (landmarks/pipeline/run.sh <name>: measure from the LiDAR, collect the Street View references, correct their cameras, render and score, and place the model on the campus map).
• How to make the next landmark: the landmark-from-code skill (.claude/skills/landmark-from-code/), with the method and every agent's brief.
• Street View is used through Google's Map Tiles API with a key of your own; its imagery stays on the computer that fetched it and is in no commit. The way of working follows Athenaion by Leon Lin.

Updating the data snapshot

Data is fetched once, cached as raw snapshots under data/raw/, and every later build reads the cache. data/sources.json records the URL, fetch time, size and SHA-256 of each snapshot.

The campus map's snapshots are not in the repository (see below), so a fresh clone cannot run npm run data:build, and a contributor does not fetch new ones: that would ask the University's map for data the project already holds, and give a public/data/uoa/ that differs from the published one. The maintainer rebuilds the data. A pull request that needs a rebuild changes the pipeline's code and says so in its first line (see CONTRIBUTING.md).

npm run data:fetch                                          # fetches only what is missing
npx tsx scripts/schools/uoa/fetch-mapsindoors.ts --refresh  # re-download the campus map (4 requests)
npx tsx scripts/fetch-osm.ts --refresh                      # re-download OpenStreetMap context (1 request per area, 7)
npx tsx scripts/fetch-terrain.ts --refresh                  # re-download the elevation tiles (29 at zoom 14)
npx tsx scripts/schools/uoa/fetch-calendar.ts --refresh     # re-download the Calendar pages (3 requests)
npm run data:build                                          # rebuild public/data/uoa from the snapshots
npm test                                                    # the data consistency tests must still pass

Then commit public/data/uoa/, data/sources.json and, if refreshed, data/raw/osm/ and data/raw/terrain/.

Rules the fetch scripts follow, and that any new one should too (scripts/lib/snapshot.ts enforces them): only public pages that need no login; at most one request per second; each URL fetched once and cached; an honest User-Agent that links to this repository. Raw snapshots of University data stay out of the repository (data/raw/ is gitignored); OpenStreetMap data and the LINZ elevation tiles are openly licensed and are committed.

When a new academic year starts, data/uoa/academic-calendar.json has to be re-transcribed by hand from the University Calendar's key-dates page (the source URL is in the file).

Adding the class timetable

The official source is the University's Classes API v2 (https://developer.auckland.ac.nz/prd/documentation/api-classes-v2). It returns, for each class, the meeting patterns with day, start and end time, date range and room: exactly what this site needs. It requires credentials that only someone with a University account can obtain, so this step is yours:

1. Get an API key for "Classes Api V2". Keys are not self-service: the Developer Portal dashboard (https://developer.auckland.ac.nz/prd/dashboard) sends you to the University's internal Athena tool, where the team that owns the API (the API document names Student Recruitment) approves the request. Say in the request that the class times will be republished on a public site. The portal's self-service OAuth credentials do not work for this API: a token is issued, but the class search still answers 401 (checked 2026-10-13).

2. Fetch and rebuild, with the key in the environment (never in a file that gets committed):

   UOA_API_KEY=... npx tsx scripts/schools/uoa/fetch-classes.ts
   npm run data:build
   npm test && npm run build
   

3. Commit public/data/uoa/ and push. Rooms then show what is timetabled at the selected time, buildings show a count of classes in progress, and courses become searchable.

[!WARNING] scripts/schools/uoa/fetch-classes.ts was written from the published OpenAPI document and has not been run against the live API (the project was built without a key), so the key header (apikey, which the API document does not spell out) or the paging parameters may need a small adjustment on first use. The part that turns API records into the published timetable (scripts/lib/classes.ts) is unit-tested, and the interface is tested end to end with a fixture timetable that exists only inside the test browser.

What the pipeline keeps for every class: course code and title, class type and section, day, times, date range, the original location text, and how precisely it could be located (room, floor or building). Records it cannot use are kept in an exclusions list with the reason. Staff names are never read or published.

Adding a language

1. Copy src/i18n/locales/en.json to src/i18n/locales/<code>.json (for example mi.json, zh.json).
2. Translate the values. Keep the keys and the {placeholders}. Set _meta.name to the language's own name; that is what the language menu shows.
3. That is all. The file is picked up automatically, the language appears in the menu in About, and ?lang=<code> selects it. Missing keys fall back to English.

Names in the data are language-keyed too ("name": { "en": "…" }), so a translation of a building or term name is one more key next to en (the calendar already carries a few mi names). The pipeline writes only en today because that is all the sources provide.

npm test checks that every key used in the code exists in en.json and that components contain no hard-coded interface text.

Porting to another university

Most of the code carries over to another campus; the data pipeline and a handful of Auckland-specific places do not. .claude/skills/campus-world-model/ is a Claude Code skill that takes an agent through a port: recon of the new school's public data, rebuilding the pipeline, shipping the site and verifying it live. Fork or clone this repository, open it in Claude Code and run /campus-world-model <university, campus>. Beside it, .claude/skills/street-look/ takes the walker's street-level look to a world model that lacks it, or tunes it for another city: /street-look bring or /street-look tune.

Contributing

Reports and fixes are welcome: see CONTRIBUTING.md. The rules every change keeps and the checks it passes are written for people and for coding agents alike, in AGENTS.md and docs/standards.md.

How it is put together

scripts/            data pipeline (Node + TypeScript, run with tsx)
  lib/snapshot.ts     polite, cache-first fetcher
  lib/context.ts      the surroundings of each area, from OpenStreetMap (every school)
  lib/terrain.ts      the ground under each area, from the elevation tiles (every school)
  fetch-osm.ts, fetch-terrain.ts   the shared fetches, over the school's areas
  schools/uoa/        the University of Auckland's own fetch scripts (campus map, Calendar,
                      classes), its campus adapter (mapsindoors.ts) and its subject list
  build-data.ts       snapshots -> public/data/uoa/*.json
  check-pr-form.ts    CI: a pull request says how it was made
  school-list.ts      CI: reads and checks .github/schools.json, the schools the site holds
  walker/             the walker's controller, CLI and MCP server (npm run walker), and the
                      script that makes its figure
data/
  sources.json        provenance of every snapshot
  uoa/academic-calendar.json   hand-transcribed key dates
  raw/                snapshots (gitignored except OpenStreetMap and the elevation tiles)
public/data/uoa/    published data: campus, context, rooms-index, subjects, calendar, schedule, meta,
                    and plans/ with one floor-plan file per building
public/models/      the walker's figure (CC0)
src/shared/         pure logic shared by pipeline and app: geometry, room codes, Auckland time,
                    calendar, schedule, search
src/scene/          three.js scene
src/app/            state, URL, data loading, translations
src/ui/             Preact components and styles
src/i18n/locales/   interface text
tests/              unit tests (Vitest)
e2e/                browser tests (Playwright), numbered after the acceptance criteria

Accuracy is labelled wherever it matters. A room drawn with an outline comes from the campus map. A room shown as a pin is known to the campus map by its code; the pin stands on the shape that the floor plan draws for it, which is outlined. The shapes carry the campus map's type for each, in its own words, and no code or name. A room that the map does not know is placed on a floor inferred from its number and marked as inferred. A building's shape comes from the campus map's outline of each floor, with floors that are alike within a twentieth of their area drawn as one block; its height is an estimate of 3.6 m a floor, and its windows and roof rims are drawn by rule.

Data sources and licences

Source	Used for	Terms
University of Auckland campus maps (the public map, served by MapsIndoors)	Buildings, floors, room codes, room types and positions	© University of Auckland. Only derived facts are published here: learning spaces with their names and simplified outlines; other searchable rooms as a code, floor and point; every other space as a simplified outline with the map's type for it and no code or name. Floors with student accommodation, childcare centres and single parking bays are left out, and a room the map marks as not searchable is never named.
OpenStreetMap via the Overpass API	Surrounding buildings, streets, paths, parks, trees	© OpenStreetMap contributors, Open Database License 1.0. The snapshot in data/raw/osm/ and the derived public/data/uoa/context.json are under the same licence.
University of Auckland Calendar 2026	Semester dates, breaks, holidays; list of subject codes	© University of Auckland. Dates and codes are quoted as facts with a link to the source.
LINZ 8 m Digital Elevation Model (2012), via the AWS Terrain Tiles	The height of the ground	New Zealand terrain data © Land Information New Zealand and the New Zealand Government, CC BY 3.0 NZ. The tiles also hold SRTM and GMTED2010 data (U.S. Geological Survey) and ETOPO1 (U.S. NOAA) at the coast. The tiles in data/raw/terrain/ and the derived public/data/uoa/terrain/ are under the same terms.
Auckland Part 1 LiDAR point cloud (2024), tiles , captured by Woolpert for Regional Software Holdings Limited and distributed by OpenTopography and LINZ

AUT World Model (/aut/) uses four of these sources and nothing of AUT's own. OpenStreetMap gives the buildings of AUT's City, North and South campuses, picked by their operator tag and AUT's building code, with their names, codes, outlines and, where tagged, heights, and the city round them; public/data/aut/campus.json is derived from the committed snapshots of its three areas (the City Campus from the same snapshot as the University's), context.json from that one, and context/north.json and context/south.json from the snapshots of those two campuses, all under the ODbL. The LINZ 8 m elevation model gives the ground (public/data/aut/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the City Campus, and Quaternius's figure is the walker's body. AUT's rooms, class times, semester dates and course list wait for AUT's permission, and its floor plans for a written agreement with MazeMap: see docs/schools/aut.md.

VUW World Model (/vuw/) uses the same four sources and nothing of VUW's own. OpenStreetMap gives the buildings of the Kelburn, Pipitea and Te Aro campuses, picked by their operator and building=university tags, with their names, outlines and, where tagged, heights, and the streets round them, from the snapshots of its three areas, data/raw/osm/context-vuw-kelburn.json, context-vuw-pipitea.json and context-vuw-te-aro.json; public/data/vuw/campus.json, context.json and context/ are under the ODbL. The LINZ 8 m elevation model gives the ground (public/data/vuw/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the Kelburn Campus, and Quaternius's figure is the walker's body. VUW's room list and course list wait for the maintainer's reading of VUW's copyright terms, and its class times and key dates sit in parts of its site that its robots.txt disallows: see docs/schools/vuw.md.

Otago World Model (/otago/) uses the same four sources and nothing of the University of Otago's own. OpenStreetMap gives the buildings of the Dunedin and Christchurch campuses, picked by their operator and building=university tags, with their names, outlines and, where tagged, heights, and the streets round them, from the snapshots of its two areas (data/raw/osm/context-otago-dunedin.json and context-otago-christchurch.json); public/data/otago/campus.json, context.json and context/ are under the ODbL. The LINZ 8 m elevation model gives the ground (public/data/otago/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the Dunedin campus, and Quaternius's figure is the walker's body. Otago's campus map, room list, key dates and paper pages are on its main website, which refused the project's first request to it; its class times answer only on its own network: see docs/schools/otago.md.

UC World Model (/uc/) uses the same four sources and nothing of the University of Canterbury's own. OpenStreetMap gives the buildings of the Ilam and Dovedale campuses and of the Arts Centre city location, picked by their operator and building=university tags, with their names, outlines and, where tagged, heights, and the streets round them, from the snapshots of its three areas (data/raw/osm/context-uc-ilam.json, context-uc-dovedale.json and context-uc-arts-centre.json); public/data/uc/campus.json, context.json and context/ are under the ODbL. The LINZ 8 m elevation model gives the ground (public/data/uc/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the Ilam campus, and Quaternius's figure is the walker's body. UC's copyright page asks for its written permission before material on its webpages is copied, which holds back its rooms, key dates and course list, and its timetable planner is on a site whose robots.txt disallows every automated visitor: see docs/schools/uc.md.

Massey World Model (/massey/) uses the same four sources and nothing of Massey University's own. OpenStreetMap gives the buildings of the Manawatū, Auckland and Wellington campuses, picked by their operator, building=university and amenity=university tags, or a name or an address that says Massey, with their names, outlines and, where tagged, heights, and the streets round them, from the snapshots of its three areas (data/raw/osm/context-massey-manawatu.json, context-massey-auckland.json and context-massey-wellington.json); public/data/massey/campus.json, context.json and context/ are under the ODbL. The LINZ 8 m elevation model gives the ground (public/data/massey/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the Manawatū campus, and Quaternius's figure is the walker's body. Massey publishes no room outlines the site could find, and its class timetable is behind a sign-in; its key dates and course list are open for non-commercial use but are not read yet: see docs/schools/massey.md.

Waikato World Model (/waikato/) uses the same four sources and one of the University of Waikato's own: its Hamilton campus map. OpenStreetMap gives the buildings of the Hamilton campus, with their names, outlines and, where tagged, heights, and the streets round them, from the snapshot of its one area (data/raw/osm/context-waikato-hamilton.json); public/data/waikato/campus.json and context.json are under the ODbL. Which of those buildings are the University's comes from the University's Hamilton campus map (version Q1 2025), which the site's maintainer downloaded in their own browser and gave the project on 2026-10-28: a table in src/schools/waikato/school.ts ties each building to the map's block, and the 17 buildings OpenStreetMap leaves unnamed are called by the map's block codes. The map is the University of Waikato's and is not in the repository; About lists it with the other sources. The LINZ 8 m elevation model gives the ground (public/data/waikato/terrain/), under CC BY 3.0 NZ. Open-Meteo gives the weather at the Hamilton campus, and Quaternius's figure is the walker's body. Waikato's main website, which holds its campus maps, key dates and paper pages, is closed to this project by its robots.txt, so no agent of the project reads it: see docs/schools/waikato.md.

Snapshot dates are shown in the site's About panel and recorded in public/data/uoa/meta.json, public/data/aut/meta.json, public/data/vuw/meta.json, public/data/otago/meta.json, public/data/uc/meta.json, public/data/massey/meta.json and public/data/waikato/meta.json.

The code is under the MIT licence (see LICENSE). That licence does not cover the data above; NOTICE says so too, beside the licence.

If you represent the University and would like something changed or removed, please open an issue.

Known limitations

• No class times until a timetable snapshot is added (see above).
• A timetabled class is not the same as an occupied room.
• Building heights are estimated and windows are drawn by rule. The ground's heights come from the LINZ 8 m elevation model, drawn from contours: smooth, peaks a few metres low. A building stands at the ground under its middle, with a footing of up to 4 m on a slope. The Clock Tower (105 and 105S), Old Government House (102), the General Library (109), the range of Old Choral Hall (104) on Symonds Street, Maclaurin Chapel (107), the Clock Tower East Wing (119), 9 Eden Crescent (801) and 15 Eden Crescent (802) are the exceptions: their models are drawn to sizes measured from LiDAR. The measured massings (About lists them) are measured too: their heights and roofs come from the LiDAR, their walls and windows are drawn by rule. Three of them, Alfred Nathan House among them, are buildings planned as landmarks: until their landmarks come they are drawn as plainly as the others, with no detail of their own.
• Only Old Choral Hall's range on Symonds Street is drawn in detail. Every source shows the building during building works: the LiDAR of May 2024 has scaffold decks over its north-east blocks, and the newest Street View panoramas show them wrapped. Only the south-west range and its two pavilions are seen as they stand, before the works (2022-09) and after them (2024-10). The rest of the building keeps the map's estimated block (two storeys, flat-topped, plain walls) until the works end, so it is lower than the building is.
• Old Government House is seen in only two photographs. Trees hide it from the other Street View panoramas near it, and the two see the upper storey of its south-east end. Its sizes and roofs come from the LiDAR. What no photograph shows is drawn plainly or after the parts the photographs do show: most of the garden front (its windows repeat the two that are seen), the porch, the north-west wing, the backs of the main block, the north-west lobby, the low south-east range, the north-west balcony (like the south-east one) and the ranges round the courtyard behind.
• The General Library is drawn plain, or after the faces they do show, where its photographs do not show it, and simpler than them where they do. None of its twelve Street View panoramas shows the main block's back on the service lane, the low block's north-east face or either end of the back wing: the back wing's north-west end is drawn with the fins of its north-east face, the others plain. At the south-east end, the two-storey curtain wall over the entrance is drawn as the slab's ribbon windows, the dark stone cladding and the louvre of the ground floor as glazing between the grey piers, and the north-west end's stone panel is drawn white.
• Maclaurin Chapel is seen from Princes Street only, through trees. Five Street View panoramas see its street side; its heights come from the LiDAR. What no source shows is drawn plainly: the small room beyond the bay, which a tree hides in the LiDAR and in every photograph (white walls on the campus map's floor outline, a flat roof at the height the LiDAR finds under the tree), the colour of the bay's and that room's roofs (pale), and the faces away from the street (plain white walls). The row of skylights in the hall's far roof, which the aerial photograph shows, is not drawn.

Inspiration

This project was inspired by the Tsinghua World Model, an independent student project that lets you walk through Tsinghua University's campus and find courses and classrooms in it. The idea of a campus you can step into comes from there. The code, models and data here were made separately; nothing is copied from that site.

Star history