Make the map of science — and its blank spaces — visible.

Mission. COMPASS ID, Inc. is a 501(c)(3) nonprofit that uses AI to classify the world’s scientific literature against controlled, expert-authored vocabularies. We build a free, open-access map of what science knows and what it has not yet studied, and we make research gaps measurable so that knowledge, funding, and conservation effort reach the places and questions that need them most. We are a nonprofit because that map should be a shared public good, open to everyone, not a product sold to a few.

Why this matters

Millions of research papers record what we have learned about species, ecosystems, diseases, places, and the policies meant to protect them. But there is no index of that knowledge — no way to ask, across the whole record, where research is concentrated and where it is entirely absent. As a result, effort clusters on the well-studied while urgent questions with no indexed research stay invisible.

There is a specific reason for that. The metadata every paper carries — its DOI, its authors and their ORCID identifiers, their institutions, the journal — has no field for the place the study actually studied. A screen of Ghanaian plants used against malaria is filed under the United Kingdom, because that is where its authors worked. So when anyone counts research “about” a country, they are usually counting where the researchers were employed — which in drug discovery means crediting the laboratory and losing the forest. And “where” is not only geography: where in the body, where in the cell, where in the genome are missing from the record in exactly the same way. Whatever coordinate system a field cares about, that location has to be read out of the paper, because nothing hands it to you. We wrote about what this hides →

That index is possible because of one trick: have AI read every paper and attach the same accurate, structured metadata to each one — what was studied, and where. Once every paper carries that same structure, the pile of literature becomes a database. It can be queried in plain language instead of just keyword-searched, and because location is one of the fields, it can be mapped, so we can see directly where research has and has not happened. That location is the field the record never had: we read each study site out of the paper’s own text, resolve it to coordinates through a gazetteer (a dictionary of place names with their locations), and set it against the countries where the species is known to occur. Both steps have limits we state wherever the figures appear: the coordinates are machine-read and checked on a sample, and the ranges are country lists, not maps of where a species actually grows. Reasoning with those places, which species have been studied inside their own range, where research clusters and where it is absent, how far from a city or a station the work happens, where the environment is changing fastest with the least research, and, in beta, how much research a place would be expected to have given what we know about it, is what we mean by spatial intelligence.

For biodiversity and conservation, the cost is real: species can go extinct before they are studied, and funders cannot easily see which gaps their investment would close. COMPASS ID makes “what has not been studied?” a question you can answer from the data.

What we do — our programs & services

1. The COMPASS ID platform · IN DEVELOPMENT

A free, searchable, spatially-explicit map of the scientific literature. Papers are classified against a controlled codebook of 46 dimensions shared across domains plus domain-specific extensions — species, region, method, threat, evidence stage, policy framework, and more — so researchers can search by meaning, align work to funding frameworks, and see where research exists and where it is missing.

2. Open research atlases

Interactive, openly available case studies that apply the method to a whole family or domain — Apocynaceae (medicinal plants), the Brazil-nut family across the Amazon, global desertification and drylands, and the fungal kingdom. Each atlas shows where research is thick, where it is absent, and how it lines up against real-world need. Explore the atlases →

3. Open methodology

We publish our methods and measure our accuracy rather than assuming it. The entities we tag come from external references (IUCN, GBIF, MeSH), each tag cites a quote from the paper, and a separate model reviews each result. The tagging pipeline itself has not yet been peer reviewed; related work on the conservation metadata gap and an earlier classifier appeared in Environmental Research Letters and Ecological Informatics.

4. Research identity for scientists

Researchers can create a COMPASS ID, bring in their work by DOI or ORCID, and see their own research footprint mapped across species, ecosystems, geographies, and policy frameworks — and how it connects to the wider field.

Who we serve

• Researchers — find where the questions with no indexed research are, and position new work where it adds most.
• Funders & philanthropies — see where investments overlap and where gaps remain, and measure the coverage their funding creates.
• Institutions & agencies — direct scarce resources toward under-studied species, regions, and questions.
• The public — a freely accessible view of what science knows about the living world.

How the tagging works

• The species, places, and other entities we tag come from established reference databases — IUCN, GBIF, and MeSH — not from lists we make ourselves.
• Each tag has to cite a specific passage in the paper. If there is no supporting quote, the tag is not kept.
• We check accuracy by having a separate model review each tag without seeing the first answer, and we report the results.
• Our atlases and data are openly available. We start with conservation, and the same method applies to other fields.