Atlas · Niger
/
Methodologies
METHODOLOGIES
How recommendations and briefs are generated
A deterministic rules engine, not a language model.
Method
Each recommendation is produced by a rule that reads Atlas' own data, stores the evidence numbers and the calculation used, and carries a confidence label (CALCULATED, INFERRED, ASSUMED). Re-running replaces the previous output. Briefs are assembled from the same recommendations.
Limits. Rules encode expert judgement about what is worth flagging; they are transparent but not exhaustive.
Related reading
More in Methodologies
Still have a question?
The AI analyst answers from these articles and from the data.
Ask Atlas