Australia's Big Things

A dataset and interactive map of Australia's Big Things — the giant novelty roadside sculptures that dot the continent, from the Big Banana at Coffs Harbour to a 250-metre walk-through earthworm that stood on the Bass Highway until 2020.

https://bigthings.downunder.blue

Zero runtime dependencies. One canonical dataset. Every pin says how precisely it is placed.

npm test          # 171 tests
npm run build     # rebuild the dataset and the web app from the cached sources
npm run serve     # open the map at http://localhost:8099
npm run serve:about     # the About/landing page
npm run build:public    # package the map into public/ for static hosting
npm run serve:public    # preview public/ exactly as a static host would serve it
npm run admin     # local-only admin UI for manual corrections — see docs/ADMIN.md
node src/cli.js prawn

---

What's in here

Path	What it is
data/bigthings.json	The canonical dataset. Every big thing, with coordinates, precision, size, era, status and per-row sources.
data/overrides.json	Hand-curated corrections and additions. Each carries a reason and a source.
web/index.html	The generated map app. Not hand-edited.
web/about.html	The generated About/landing page.
web/about-template.html	Its shell. Every figure in it is injected at build time.
web/img/	284 Wikimedia Commons photographs, vendored with their licences.
web/vendor/	Leaflet, the fonts, and a vector Australia outline — all self-hosted.
web/template.html	The app shell web/index.html is generated from.
public/	The site, repackaged for static hosting. Generated by npm run build:public, never hand-edited.
data/site.json	Canonical domain and share image for SEO/AEO tags — see below.
data/image-credits.json	Per-photo photographer, licence and checksum.
src/	The pipeline: fetch → extract → normalise → build → generate.
cache/	Unmodified snapshots of every upstream source, plus CHECKSUMS.txt.
test/	The test suite, including a map-vs-CLI equivalence diff.
docs/	BUILDING.md · DATA.md · RECREATING.md · ADMIN.md

---

The honest headline

Wikipedia's article cites 1,075 big things in Australia as of 2022, sourced to Amy Clarke's research in the Journal of Australian Studies. That is a census figure, not a list — the article's own tables enumerate far fewer, and popular round-ups quote anything from "more than 150" to "around 600" depending on how strictly they define a big thing.

This project ships 477 big things: every one that two openly-licensed lists actually name, cross-checked against OpenStreetMap and independent reporting, plus 100 more mined from two community catalogues that enumerate far more than Wikipedia does. So the map's own framing is "477 mapped of a claimed 1,075" rather than a pretence of completeness.

The second honesty problem is coordinates. Wikipedia's table gives a town, not a point. A map that silently drops hundreds of town centroids on the continent is lying about where the Big Potato is. So every record carries a precision field, the app renders town-level pins with a dashed border and a 📍 town-level pin badge, and there is a filter to hide them entirely.

Precision	Meaning	Count
exact-article	The sculpture's own Wikipedia article has coordinates	20
exact-wikivoyage	A Wikivoyage marker for this name in this state	27
exact-osm	A name-matched OpenStreetMap feature near the right town	79
exact-inline	A {{coord}} template in the source table	6
exact-verified	Independently researched, or published by a community catalogue	181
town	The town's coordinates — not the sculpture's	164

313 of 477 (66%) are located to the actual sculpture; the remaining 164 sit at their town's centre. Every card states which it is, in plain words: "How this pin was placed: core name 'swan' matches exactly, 0.6 km from the town point", or "This pin sits at the centre of Monash, not on the sculpture — the sources give a town, not a point."

Run node src/cli.js --stats for the current breakdown.

---

Two pages, one build each, no hand-typed numbers

There are two pages — a landing/About page and the map — one template each, web/about-template.html and web/template.html, generating web/about.html and web/index.html. No CDN, no hotlinking: photos come from web/img/, Leaflet and the fonts from web/vendor/, all served from this domain rather than someone else's.

The About page is worth a note: every figure on it is read out of data/ at build time. The headline counts, the precision table, the state and category bars, the superlatives in prose, the correction cards and the photo credits are all generated. A landing page with hand-typed statistics goes stale the first time the dataset changes and then quietly misrepresents the thing it is introducing — so test/about.test.js asserts every number on the page against the dataset, and the precision tiers must sum to the total.

It also picks its correction cards by substance rather than alphabetically, and dedupes records that share one override, so the Nullarbor bull leads instead of three near-identical Sapphire entries.

The basemap: a coastline, then tiles

Loading OpenStreetMap tile imagery for the whole continent at once is wasteful, and at that scale it looks worse than a simple map — so the app draws a vector Natural Earth coastline at the continent view, and swaps to real OpenStreetMap tiles once you zoom in past state level. standalone.test.js asserts the swap is gated on zoom, not loaded unconditionally.

Every other asset stays local at every zoom level: photos, fonts and Leaflet never leave this domain. A test asserts the page loads no external image, font or script — tiles, fetched live per view once you're zoomed in, are the one and only intentional exception.

The map

Generated from the dataset, no build step and no framework beyond Leaflet.

• Clustered pins across the continent, coloured and emoji-glyphed by category
• Filters for state, category, era, plus two honesty switches — hide the demolished ones, or show only pins located to the actual sculpture
• Detail cards with a Wikimedia Commons photo, size, year, the source's own prose, a provenance list, and directions
• Road trip builder — add stops, hit Optimise for a nearest-neighbour route, get total distance and a Google Maps link for the whole chain
• Superlatives — tallest, longest, oldest, newest, most-copied name, and which state has lost the most
• 🎲 Spin & drive — a random big thing from whatever is currently filtered
• Place-name labels — an opt-in switch under Map display; labels stay off until you're zoomed in far enough to read one without 477 towns overlapping
• Direct links — the address bar always reflects what's on screen: open a card and the URL becomes #thing=<id>, filter by state and it becomes #state=NSW (or NSW,QLD). Load either and the map jumps straight there. No history entry per pin click — it replaces, not pushes.
• Works on a phone: the sidebar and cards become bottom sheets, nothing is clipped, no functionality is desktop-only

The CLI

The same dataset, queryable from a terminal. src/query.js is the single definition of "what matches"; the map mirrors it and test/equivalence.test.js diffs the two so they cannot drift.

node src/cli.js prawn                                    # free-text search
node src/cli.js --state QLD --category seafood
node src/cli.js --near -33.87,151.21 --within 300 --sort distance
node src/cli.js --gone --format csv > lost-big-things.csv
node src/cli.js --format geojson > bigthings.geojson
node src/cli.js --stats

Exit codes: 0 results found · 1 no results · 2 usage error.

---

Corrections we made to the sources

Open data is not the same as correct data. Each of these is recorded in data/overrides.json with a reason and a source, and surfaced in the app on the affected card:

• Big Triceratops was filed under New South Wales with the town spelt "Ballendean". Ballandean is in Queensland's Granite Belt. State and spelling corrected.
• The Big Pineapple (Woombye) was being read as demolished. The trigger was a subordinate clause about a Hawaiian water tower being dismantled in 1993. The status parser now only reads main clauses; the pineapple is heritage-listed and reopened in June 2024.
• The Big Prawn (Ballina) was being read as replaced. Bunnings refurbished it in situ after the 2009 demolition vote — refurbishment is no longer treated as a lifecycle change.
• The Giant Worm is 250 m long, not 250 m tall. The size parser now refuses to claim a height from ambiguous dimensions rather than inventing a world record.
• Larry the Lobster had its location as "Rosetown"; it stands at Kingston SE.
• The Big Oyster (Taree, 1990) and Big Ned Kelly (Glenrowan, 1992) had empty year cells.
• Big Lawnmower and Big Chainsaw at Beerwah are absent from both wiki lists and were added from independent reporting.
• Fergus the Bull was pinned on the Nullarbor Plain, 1,200 km from where it stands. Its Location cell reads "Forrest Highway, just north of Bunbury"; the geocoder stripped the road type and matched Forrest, WA — a real locality near the SA border. The geocoder now tries road-derived place names last, so the whole class of error is closed.
• The Big Redback read as standing despite having been disassembled, because mainClause() cut the sentence at "has since moved" — treating an adverb as a subordinating conjunction and discarding the only word that said it was gone.
• The Anmatjere Man appeared twice, once under Wikivoyage's name for it ("Big Aboriginal Hunter"). The duplicate is removed and the survivor moved from the Anmatjere region centroid to the Aileron Roadhouse where it actually stands.

Every one of those lives in data/overrides.json, applied after harvesting so it always wins and always survives the next rebuild — unlike data/bigthings.json itself, which is a build artifact and gets fully regenerated every time. npm run admin is a local-only web UI (http://127.0.0.1:8098, never deployed) for making your own: fix a field, pin a better coordinate, or attach a photo that isn't on Wikimedia Commons. See docs/ADMIN.md.

---

Photographs and their licences

300 photographs ship with the map: 278 vendored from Wikimedia Commons at 480px wide, and 22 added by hand (data/custom-photos.json, see docs/ADMIN.md). The licence mix:

• CC BY-SA 4.0 — 150 photos
• CC BY-SA 3.0 — 65 photos
• Public domain — 31 photos
• CC BY 3.0 — 16 photos
• CC0 — 11 photos
• CC BY 4.0 — 6 photos
• CC BY 2.0 — 5 photos
• CC BY 2.5 au — 4 photos
• CC BY-SA 2.0 — 3 photos
• CC BY 2.5 — 3 photos
• CC-BY-SA-3.0 — 3 photos
• CC BY-SA 2.5 — 1 photo
• PD — 1 photo
• CC BY-SA 3.0 au — 1 photo

Attribution is a licence condition, not a courtesy. Most of these are CC BY or CC BY-SA, which require naming the author and the licence — and for several versions this app displayed them bare, which was a breach. Every photo now carries a credit line on its card, and docs/IMAGE-CREDITS.md (and credits.html on the site) lists every one, Commons and custom, with photographer, licence and a link to the file page.

The photographs are not covered by this repository's licence. Each stays under the licence its author chose; the dataset is CC BY-SA 4.0. data/image-credits.json records the licence, the author and a SHA-256 of every local copy, and a test fails if any file drifts from its checksum or lacks an author.

Every referenced photo was vendored successfully.

SEO and AEO

The map is a client-rendered single page, so without help a crawler that doesn't run its JS sees a title and a description and nothing else. src/seo.js closes that gap on both pages, driven by data/site.json (set siteUrl there if the domain ever changes):

• Canonical link, Open Graph and Twitter Card tags, built from the same title/description strings used in <title>/<meta name="description"> — never typed twice.
• WebSite + Dataset + ItemList JSON-LD on the map page: every one of the 477 things' name, coordinates, address, photo and (where it exists) Wikipedia link, readable straight out of the raw HTML — including its #thing=<id> deep link, the same one the app's own direct-link feature understands (see "Direct links" above).
• FAQPage + BreadcrumbList JSON-LD on the about page, reusing the same computed superlatives (oldest, newest, biggest, tallest, the lost count) that already appear in the visible prose, so the two can't disagree.
• public/robots.txt and public/sitemap.xml, generated at packaging time (below).

Details in docs/BUILDING.md.

Deploying public/

npm run build:public packages the map (already self-hosted and zero-network, per standalone.test.js) as a plain static site — index.html, about.html, credits.html, 404.html, img/, vendor/, robots.txt, sitemap.xml — under public/, ready to point any static host at. For Cloudflare Pages:

npm run build:public
npx wrangler pages deploy public --project-name=bigthings-downunder-blue

Or connect the repo in the Cloudflare dashboard with build command npm run build:public and output directory public. No environment variables, functions, or KV/D1 bindings are needed — it's static files only. public/_headers sets long-lived caching for img/ and vendor/; Pages picks that file up automatically.

Sources and licensing

Source	Used for	Licence
Wikipedia — Big things (Australia)	The master list: names, towns, years, sizes, notes, images	CC BY-SA 4.0
Wikivoyage — Australia's big things	Surveyed marker coordinates, travel-guide voice, extra entries	CC BY-SA 4.0
OpenStreetMap (Overpass)	Independent coordinate cross-check; basemap tiles	ODbL 1.0
Wikimedia Commons	Photographs, vendored locally with credit	various free licences
MediaWiki API prop=coordinates	Article and town coordinates	CC BY-SA 4.0
Land of the Bigs	Big things absent from the wikis: name, town, mapped coordinate, subject category	facts only, cited per row
Aussie Big Things Passport	Same, cross-checked	facts only, cited per row

From the two community catalogues we take only facts — a name, a place, a coordinate, a subject category — which are not copyrightable, and cite the page each came from. No prose is copied: a discovered record's notes stay empty rather than lifting someone else's writing. Both sites' robots.txt permits crawling, and the harvester serialises requests with a delay.

This repository's dataset is published under CC BY-SA 4.0 to match its upstream sources. Attribution appears in the app footer and in data/bigthings.json's meta.sources.

Basemap tiles come from OpenStreetMap's public tile servers — fine for local use and modest traffic, but read the tile usage policy before deploying this anywhere busy, and switch to a paid tile provider if you do.