A working map of documented archaeological sites across Polynesia, Melanesia, and Micronesia — all three now genuinely represented, not a Polynesia atlas with a few other regions bolted on — plus tools to measure real distances between sites and test whether their layout looks any different from chance. Data is sourced from the National Register of Historic Places, UNESCO documentation, and published surveys — see each site's source note. The Pacific overview further down the page is separate scaffolding specifically for the wider Polynesian triangle — every named corner is now backed by real documented sites, with the underlying mechanism kept ready for any further waypoints that get added later.
How this works
Click any marker on the maps below to pull up what's actually documented about that site. A 📜 next to a site's name means it also has a recognized legend or oral tradition attached — shown separately from the archaeological facts, never blended in. A 🐇 means the site also has a collapsed-by-default "Down the Rabbit Hole" section — popular fringe theories people tell about the place, clearly disclaimed as entertainment, not archaeology.
Click a second (or third, fourth...) marker to compare real distances, compass directions, and possible celestial alignments between them, in the panel underneath the maps.
Hit Run test in the "Test for a non-random pattern" card to check whether the layout looks any different from 1,000 randomly scattered sets of the same size — that's how you tell a real pattern from wishful thinking. Scope it to one island or all of them combined.
Scroll down for a Timeline of every datable site, a Glossary for unfamiliar terms, and a data confidence dashboard showing what's solid versus still flagged for verification.
Not sure where to start? Hit 🎲 Surprise me above the map for a random site.
Got a hunch about somewhere specific? Log it under "Your own leads" — kept separate from the verified record so the two never get blended together.
Data model: flat SITES array, each entry tagged with island, type, a precision flag (exact = published monument coordinate, approx = general area, deliberately not pinpointed), yearStart/yearEnd (nullable, hand-derived from the era text for timeline placement only), yearUncertain, and confidenceFlag. Optional elevationM holds a site's elevation above sea level only where a cited source states it explicitly — populated for 3 of 133 sites so far; deliberately left absent rather than estimated everywhere else, since most cited sources simply don't report it. Optional lore + loreSource hold a recognized legend/oral-tradition account, separately cited from the archaeological source — only populated where a real recognized source publishes one, never invented. Optional rabbitHole holds an array of popular fringe-theory items (title, note, source, url) — collapsed by default, clearly disclaimed as entertainment, never treated as a claim.
Maps: every island now renders a real Leaflet map on free CartoDB dark-tile basemaps (OpenStreetMap data), not a schematic layout — lazy-loaded per island via IntersectionObserver so all 84 don't mount at once, with map/marker instances persisted in islandMapRegistry so clicks only restyle existing markers rather than rebuilding anything.
Distance / bearing: real haversine great-circle distance and initial compass bearing between selected sites, computed directly from each site's actual lat/lng — independent of the map display itself.
Alignment check: apparent-horizon solstice sunrise/sunset azimuth (cos A = (sin δ − sin h·sin φ) / (cos h·cos φ), δ = ±23.44°, h = −0.833° to account for standard atmospheric refraction + solar semi-diameter) plus cardinal directions, matched within ±3° of the observed bearing. Still doesn't model per-site elevation or local horizon terrain — noted in the UI caveat.
Pattern test: observed mean nearest-neighbor distance vs. 1,000 Monte Carlo draws inside each region's approximate land extent (ISLAND_BOUNDS), now scopable to a single island or all combined via the region select. Reported as a percentile rank, not a p-value — treat a hit as a lead worth chasing, not a proof.
Extending it: add a region by adding entries to SITES plus a matching box in ISLAND_BOUNDS. The Pacific overview's TRIANGLE_NODES is a separate, lighter reference layer with no site-level claims yet. Add real terms to GLOSSARY and real entries to CHANGELOG as the atlas actually changes.
Export: CSV/GeoJSON generated client-side from SITES via Blob + anchor download — no server round-trip.
Leads: stored per-visitor via window.storage, not shared or visible to anyone else, and excluded from data exports.
Sites — click to select, click again to deselect
Site record
Click a site on the map to see what's documented about it.
All sites ()
Measure between selected sites
Select two or more sites on the map to see distances, compass bearings, and possible celestial alignments between them.
From
To
Distance
Bearing
Alignment
Alignment check: compares each bearing against solstice sunrise/sunset and cardinal directions (±3°), corrected for standard atmospheric refraction and the sun's apparent size at the horizon. Still assumes flat, unobstructed terrain — doesn't account for a specific site's actual elevation or local horizon (hills, ridgelines) blocking the view. A match is a lead worth investigating, not proof of intentional alignment — especially over these distances, where many bearings will land near a solstice azimuth by chance alone.
Test for a non-random pattern
Compares how spread out the documented sites actually are against 1,000 randomly generated site sets of the same size, placed within the same rough island extents.
Timeline
All datable sites plotted on one timeline. Bands are hand-derived from each site's own era description above for visual placement only — treat exact edges as approximate, not new claims. Sites with no usable date are listed separately below.
Data confidence dashboard
A transparency snapshot of what's solid versus what needs more verification — the same distinction the site record panel shows per site, aggregated here.
Glossary
Plain-language definitions for terms used across the site descriptions above.
Advertisement
Export data
Download the full documented dataset to run your own analysis outside this app.
Exports include every field shown in the site record panel: id, name, island, country, type, coordinates, precision, era, description, and source. "Your own leads" are not included — they're personal and stay local to your browser.
Bibliography
Every source cited above, compiled in one place.
Changelog
Pacific overview — the wider Polynesian triangle
Real coordinates for all eight named corners of the Polynesian triangle and its key waypoints — every one of them is now backed by documented sites elsewhere in this atlas, not just reference points. The filled/hollow marker distinction below stays functional for future additions: if a waypoint gets added here before it has real sites researched, it'll render hollow until it does.
Documented in this atlasReference point only — no sites loaded yet
This panel uses a flattened, whole-Pacific projection for orientation only — it is not accurate enough for the distance/bearing measurements above, which stay scoped to sites within a single island group.
A pattern worth naming — Holocene sea-level change, already documented across separate sites
Not a new discovery from a handful of elevation numbers — elevation itself is only documented for 3 of 133 sites so far, nowhere near enough to support a pattern claim on its own. This is different: a real, independently-cited geological pattern that was already sitting scattered across several unrelated site entries, pulled together here into one place.
Multiple, separately-sourced entries in this atlas independently document the same broad phenomenon: Pacific sea levels were measurably higher in the mid-Holocene than they are today, followed by a later relative fall — the well-established mid-Holocene highstand, not to be confused with present-day, human-driven sea-level rise, which runs the opposite direction. Four sites currently on record here touch on it, from entirely different research programs:
Kilu Cave, Bougainville — Ice Age sea levels low enough to join Buka, Bougainville, and several neighboring islands into one landmass.
Bourewa, Fiji — sea levels roughly 1.5 meters higher than today around 3,000 years ago, when Lapita voyagers first arrived.
Laura Village, Marshall Islands — the habitable islet itself only emerged around 2,000 years ago, as relative sea level dropped.
Seseleka Hillfort, Fiji — a further sea-level fall around 1300 CE is linked by researchers to a food crisis that pushed coastal communities uphill into fortified settlements, a pattern echoed at multiple other Fijian hillforts in this atlas.
Four sites, four separate research teams, one consistent regional signal — that's a real pattern worth naming, not four coincidences. But it should be read as a synthesis of existing, independently-published findings, not as new analysis performed on this atlas's own data. Click any site above to jump to its full record and citation.
Your own leads
A place to log a hunch, an oral-history reference, or a location worth checking — kept separate from the documented record above so the two don't get blended together.