From the lab. Written carefully.

Short essays on how Mara works, how we measure it, and the choices we make about what to publish and what to keep closed.

1. R-01

   Reasoning about an unknown sample.

   How Mara approaches a piece of malware it has never seen before: what to look at, what to suspend judgement on, and where to begin.

   8 min
2. R-02

   Attribution under uncertainty.

   Why Mara hedges on threat-actor names, and why we believe a model that hedges well is more useful than one that names confidently.

   6 min
3. R-03

   A note on evals.

   How we measure whether Mara is actually useful for the work analysts do, not the work benchmarks pretend they do.

   7 min
4. R-04

   Working in the open.

   Why our research notes, our model cards and our evals are public, and which parts of Mara we keep closed, and why.

   5 min