Posts

News, announcements, blog posts, and paper-like investigations from Dalea.

5 posts
Someone else must have run this experiment before...
FeaturedPaper · Aug 17, 2026 · @rasmus-dalea

Someone else must have run this experiment before...

They almost certainly have. The problem is that what they learned is published as guidance, while the setup itself stays trapped in their lab. Weekly Papers #4 on why scientific structure rarely travels.

Dalea is now an official community connector for Claude.
News · Jul 29, 2026 · @thomas-dalea

Dalea is now an official community connector for Claude.

The foundation of Dalea is build on many philosophies, one essential part is: a research platform shouldn't have a human version and an AI version. It should b...

Biology has no compiler
Blog · Jul 23, 2026 · @thomas-dalea

Biology has no compiler

Agentic AI arrived in the lab this year on the strength of what it did for software. But what makes coding agents work, a compiler and a test suite that tell you deterministically whether the work is right, has no equivalent in biological research. Weekly Papers #3 on why the answer to trustworthy AI in life science isn't a bigger agent, more agents, or more skills, but a deterministic substrate that audits and traces every action, whether human, AI, or programmatic, in a language built for science.

The Same Algorithm, Half the Patients
Paper · Jul 16, 2026 · @rasmus-dalea

The Same Algorithm, Half the Patients

Two health databases. Same year, same country, same conditions - and on average every condition looked 12% more prevalent in one than the other. Then someone standardised what counted as an "inpatient visit," and the gap fell to 1%. Weekly Papers #2 on why checking another group's work means rebuilding their data model, why almost everyone rebuilds it from scratch, and what it would mean to install it instead.

An ELN isn't just an ELN
Paper · Jul 10, 2026 · @rasmus-dalea

An ELN isn't just an ELN

When we share lab work, we argue about export formats — PDF, CSV, a folder of attachments. But that's the wrong question. Before asking how to export the work, ask what the work actually is: a protocol, a data model, code, queries, and charts that depend on one another. Weekly Papers #1 on why the natural unit of scientific sharing is the package, not the page.