Statistical Analysis of Ancient Monumental Site Distribution Along a Proposed Great Circle: Evidence from Five Archaeological Databases and a Hemisphere Decomposition

We present a quantitative analysis of the spatial distribution of ancient archaeological sites relative to a proposed great circle alignment (Alison, c. 2001). Using five archaeological databases — the Megalithic Portal (61,913 sites), the Pleiades Gazetteer (34,470 sites), the p3k14c radiocarbon compilation (36,693 unique sites), the Digital Atlas of the Roman Empire (29,760 sites), and Historic England (20,026 scheduled monuments as a negative control) — we test whether ancient monumental sites cluster near this great circle beyond what geography and population density predict.

After adjusting for habitability — comparing the circle against equally populated corridors — the overall site count is fully explained by the circle's path through habitable low-latitude terrain (78.5th percentile among habitability-matched circles, not significant). The aggregate alignment is geographic.

However, within those corridors, monumental sites cluster while settlements do not. On Pleiades ancient sites, monuments enrich at 5× (Z = 8.55) while settlements fall below random (Z = −1.43), producing a divergence of 9.98 Z-units. This divergence replicates across four independent databases with different classification methods, survives habitability adjustment (99.63rd percentile among 4,323 matched circles), is not produced by any of 10,000 random great circles, and is not dependent on any single continent — surviving removal of Egypt, removal of Africa, and independently reproducing in South America alone. A negative control (Historic England) confirms null where expected.

Split-sample blinded validation confirms the signal on held-out data (mean Z = 9.45, 100/100 splits). Multi-scale enrichment analysis shows peak significance at 10–20 km, with settlements never reaching significance at any scale. All primary findings survive Benjamini-Hochberg FDR correction across 42 tests. All code and data are openly available