About ResearchMap
ResearchMap reads the research papers on one subject and lists questions those papers leave open: things a paper said should be studied next that no later paper in the collection took up, limits that keep coming back, and pairs of findings that appear to disagree. Each result links to the papers and sentences behind it.
What it is not
- It is not complete. It only knows the papers in the library you have selected, so “nobody has answered this” means “no paper in this library answered it”.
- It is not peer review and not advice — medical, dietary or otherwise.
- It is not validated. We have not yet tested whether the questions it lists turn out to be ones researchers later worked on.
- It does not read every paper in full. Where no open copy exists, it reads the abstract only, and says so on the paper's page.
How it works, in five steps
- We pick one subject and collect papers on it from OpenAlex, an open catalogue of research.
- A language model reads each paper (the full text where an open copy exists, otherwise the abstract) and writes down, in a fixed format, what the paper claims, the limits it admits and what it says should be studied next. It does not rank or judge anything.
- Plain code compares those notes across papers to find unfollowed questions and recurring limits. For possible disagreements, code picks pairs of similar claims and a language model is asked, one pair at a time, whether the two conflict.
- Code orders the results by fixed rules that are published with the source code, and every result keeps a trail back to the papers.
- We check results by hand where we can and write down what we found — including the verdicts we now doubt. See Method.
The libraries
| Library | What it covers | Papers | Built |
|---|---|---|---|
| Language-model reliability | How language models express confidence, when they should refuse to answer, how their uncertainty is measured, and why they state false things as fact. | 113 | 2026-07-30 |
| Diet and all-cause mortality | What the epidemiology and clinical-trial literature says about diet — food groups, dietary patterns, and specific regimes — and mortality-related outcomes. | 101 | 2026-10-10 |
| Algorithmic fairness in machine learning | Definitions and metrics for fair machine-learning classifiers, incompatibility results between them, bias-mitigation methods, and audits of deployed systems. | 109 | 2026-10-12 |
| Social media and adolescent mental health | What studies find about young people's social-media, smartphone and screen use and their depression, anxiety, loneliness and well-being, and why they disagree. | 106 | not measured |
Libraries are built one subject at a time. Diet and all-cause mortality and Algorithmic fairness in machine learning and Social media and adolescent mental health are set up to add new papers once a week, and a week's additions are published only when every automated check passes (see Method). Language-model reliability is kept fixed.