Methodology & Findings

Can a genuinely sourced, falsifiable map of Pacific archaeological sites actually be built — and meaningfully expanded across Polynesia, Melanesia, and Micronesia — without turning into a pseudo-archaeology confirmation tool?

That's a real risk with this kind of project. Star mounds, marae alignments, and "ancient mysteries" are exactly the territory where enthusiasm outruns evidence. The test this atlas set for itself from day one: every site needs a real, named, checkable source, and every analytical tool needs to report uncertainty honestly rather than dress up a coincidence as a discovery.

Real citation or it doesn't go in — no exceptions, no filler.

• Follow the researcher, not the place name. Searching a location directly tends to surface tourism blogs and thin encyclopedia summaries. Following the actual career of a named archaeologist who worked in the region — their full bibliography, their students, their formative early digs — has consistently surfaced the real excavation reports instead. This is how sites from Hawai'i to the Trobriand Islands to Kiribati's Utiroa dig ended up in the atlas.
• Lore and fact, kept structurally separate. Recognized oral tradition or legend gets its own field, its own citation, and its own visual treatment (a dashed gold border) — it's never blended into the archaeological description, even when both exist for the same site.
• Flag the thin ones, don't hide them. Anything with weaker, tourism-only, or undated sourcing gets an explicit low-confidence flag and a plain-language explanation of why — rather than being quietly presented with the same confidence as a peer-reviewed excavation.
• Non-unfalsifiable analysis tools. The distance/bearing tool uses real great-circle geometry. The celestial alignment check accounts for atmospheric refraction and the sun's own angular size, and states plainly what it still doesn't model (site elevation, local horizon terrain). The pattern test reports a percentile rank against random simulation — explicitly framed as a lead worth chasing, never a proof.
• Non-English sources, searched directly. Several real gaps in the record only closed by searching French, Japanese, and Spanish-language heritage literature directly, rather than relying on secondary English-language coverage of it.

A concrete example of the discipline in action: after adding a real elevation field to 3 sites, the obvious next question was whether those 3 numbers showed a pattern — say, evidence of islands rising or sinking. They don't; three points isn't a dataset. But checking that question surfaced something genuinely real instead: four separate, independently-published site entries in this atlas — from four different research programs — each document the same well-established regional phenomenon, the mid-Holocene sea-level highstand followed by a later relative fall. That became a real synthesis panel in the atlas itself, explicitly labeled as connecting existing findings rather than new analysis. Same underlying question, two very different honest answers depending on what the evidence actually supported.

As of this write-up:

Every country and territory named as a target in past research sessions — including the five Micronesian nations that were once the atlas's last unfinished corner (Chuuk, the Marshall Islands, Kiribati, Nauru, and Yap proper) — now has at least one genuinely sourced entry.

These numbers will drift as the atlas keeps growing — that's the point of a living project. For the current, always-accurate count, check the live site list or its in-app changelog, rather than treating the figures above as fixed. (A static, unrefreshed count is exactly the kind of small bug this project has caught and fixed in itself before.)

Yes, with real limits — some of them permanent, not just "not done yet."

The honest answer to the original question is: yes, this can be built without becoming a confirmation tool — but only by treating "we don't know" and "the record is thin here" as acceptable, sometimes final, answers. That discipline is the actual finding of this project, as much as any single site in it.