ProAge Atlas · AI research

Intelligence,
under scrutiny.

AI can make research easier to explore. Its value depends on the evidence it uses, the questions it can answer and the corrections it still needs.

Conceptual scientific imagery Research software prototype

Inside Evidence Copilot

An answer with an
evidence trail.

Public research notes enter a bounded workflow. Every response keeps the source and its limitations close to the finding.

Source selection

Decide what
the model reads.

Search our curated library and choose up to six source notes. Their study type, findings and limitations travel with the question.

Explore the source library
Grounded generation

Keep claims
attached to context.

The assistant compares the selected notes. Each answer paragraph references that source set; unknown citation identifiers are rejected.

Ask a research question
Scientific review

Leave room
for disagreement.

Methods, populations and measurement targets can differ. The answer makes those differences visible and records what the notes cannot establish.

Understand measurement

The evaluation agenda

Useful is a
testable claim.

Our next evaluation compares assisted research tasks with a defined baseline and expert review. We have not yet published a benchmark for this assistant.

Discuss a research evaluation
Source fidelity
Does the summary retain the population, study design and outcome actually described in the note?
Unsupported claims
Does the model acknowledge insufficient evidence rather than fill a gap?
Correction effort
How much expert time is needed to turn an answer into a usable research record?
Measurement distinctions
Does it keep clinical age, epigenetic age, pace and functional outcomes separate?

Method & interpretation

What the assistant actually does.

Notes, with primary links

Curated summaries provide the working context. The assistant does not claim full-text retrieval or a systematic search. Read the original source before relying on a consequential interpretation.

Deterministic instruments

PhenoAge and study planning use published mathematical models. Generative AI does not modify their formulas, impute missing biomarkers or turn a numerical result into a diagnosis.

Open the research instruments

A research development program

Citation checks verify source identifiers and response structure. They do not establish that every interpretation is correct. Task-specific evaluation and scientific review remain necessary.

Open Institute of Age and Aging

Published methods. Traceable sources. Questions worth pursuing.

Collaborate with the laboratory