Millions of studies record what science has learned, yet there is no index of what it has and has not covered. COMPASS ID is a nonprofit building that index as a free, open-access map of the scientific record. It makes the places no research has reached visible, so effort and funding can go where they are needed. Our research atlases are open now. The full platform is still in development.
An early-stage prototype. The research atlases are open now. Researcher profiles are in beta; full search is still in development.
Illustrative example · not real data
Where the work happens
Research footprint
Policy alignment
These totals span every corpus COMPASS ID has classified, including corpora not yet published as their own atlas · each published atlas reports its own paper count, species count, tag count, and measured accuracy.
Millions of research papers hold what humanity has learned, but the literature itself is just an enormous pile of unstructured text — you can search it for words, but you can’t ask it real questions. COMPASS ID’s method is to have AI read every paper and attach the same structured facts to each one, what was studied and where, so the pile becomes a database: queryable in plain language, and mappable because location is one of the fields. We are a nonprofit because that map should be a public good, not a private product. We begin in conservation, and the same method extends to other fields.
Studies are tagged against a controlled, expert-authored codebook — so you can explore, align, and map an entire field by meaning, not keywords.
Explore a whole field as an interactive map — studies tagged, geolocated, and filterable across dozens of dimensions. Building your own is next: bring in your papers, have them tagged, and map your corpus’s coverage and gaps.
See how research lines up with the frameworks that drive funding and action — the SDGs, the UNCCD, CBD targets, CITES, Ramsar, the IUCN Red List — surfaced automatically from the literature.
Map where research actually happens — geocoded study sites, not author affiliations — and where it does not. The gaps are measured from the data, not estimated.
Our Apocynaceae atlas covers tens of thousands of indexed studies of the plant family that gave medicine vincristine and reserpine. Study sites are geocoded from the text — not author affiliations — so the map shows where the science actually happens. Yet most of the family’s species have no indexed research, and the unstudied ones sit far from where the research is.
of species in the Apocynaceae — the family that gave medicine vincristine and reserpine — have no indexed research.
Every accepted species is loaded from a taxonomic authority and matched against the indexed literature, so a species with no linked papers is a measured result, not an estimate.
The tags drive the maps, the filters, and the gap counts, so they need to be accurate. Each one is tied to the paper’s own text, checked by another model, and measured against real data.
The species and places we tag come from IUCN, GBIF, and MeSH, not from lists we make ourselves.
Each tag cites a specific passage in the paper. Without a supporting quote, it is not kept.
A second model reviews each tag against the same rubric, and a third checks the result.
Against a held-out expert gold standard, tag precision runs 81–86% across published corpora (recall is deliberately conservative, so counts are undercounts, not overclaims). Geocoding and species-link accuracy are measured and reported separately. Each atlas states its own precision, recall, method, and date.
Now in beta · coverage is partial · a blank region means “no indexed study found,” never “no research exists.”
Explore our open research atlases and the interactive map, or learn how the method works. More atlases are on the way — if you are an organization or researcher who wants one for your field, region, or species group, request it.
