Lost Sites
WithheldCandidate archaeological sites triangulated from colonial-era written sources, absent from modern national registers.
Gazetteers, missionary memoirs and expedition reports describe structures that no modern archaeological register lists. Each candidate here is one coordinate backed by one firsthand source I judged trustworthy, factual and locatable, with satellite imagery used only as optional soft corroboration and never as the primary evidence. The ranking is deliberately separated from the verification: it scores how much it would matter if a site is what the source says, not how confident I am that it is. A single-witness stone complex in Central Africa outranks a triple-attested cairn, because if it holds it overturns more.
What exists54 distinct candidates ranked into tiers, 975 notes, 2,432 images. Coordinates are withheld from this site and will stay withheld. Publishing them would function as a target list for looters, and nothing about my confidence in the sources justifies that risk.
LiDAR Naturalist
ActiveReprocessing old Mesoamerican LiDAR archives with archaeological classification rather than forestry classification.
Valeriana was found in 2024 inside a 2013 carbon-inventory scan that nobody had reprocessed. The survey was flown to count trees, so the ground was never classified for archaeology and the features were simply never rendered. That is a repeatable situation rather than a one-off, and the pipeline here takes obscure or barely-digitized LiDAR archives and reruns them with ground classification and visualization aimed at mounds, causeways and reservoirs.
What existsRan on real public surveys including AMIGACarb Yucatan Norte 2013 and the OpenTopography Aguada Fenix collection. Produces hillshade derivatives and labelled candidate crops. 43 modules with Docker and PDAL. Nothing has been checked against a published site database, so the candidates are candidates only.
Congo Basin Engine
ActiveGrid-and-score targeting across the Congo Basin, rendered as interactive maps.
A scoring pass over gridded data layers for five city regions and a continental view, output as slippy maps I can actually pan around rather than a table of coordinates. Built to sit upstream of Lost Sites: this narrows where to read, rather than deciding what is there.
What existsSix interactive maps and matching target sets exist and open in a browser. There is no README, so the framing above is reconstructed from the code rather than from anything I wrote at the time.
Lodestone-Atlas
ActiveTesting whether prehistoric ritual sites sit on magnetic anomalies detectable in public data.
A seven-stage pipeline that ingests magnetic, elevation and optical data, subtracts the IGRF reference field, computes analytic signal and tilt derivative, filters on geometry, and exports ranked candidates as KML and GeoJSON. The hypothesis is old and not mine, and the honest position is that it is unproven. What I wanted to know was whether it could be tested at all using only public data.
What existsExecuted end to end on two regions, Salisbury Plain and AlUla Khaybar, producing processed rasters, ranked candidates and per-region reports. Never committed to version control, so there is no history behind the current state. AlUla candidates carry the same withholding rule as Lost Sites.
he premise across all four of these is the same. A great deal was observed, described and printed in the nineteenth and early twentieth centuries by people who had no way to record where they were standing, and a great deal was scanned in the last twenty years by people looking for something else entirely. Both are archives. Neither has been read properly.
