In the last section we looked at who published the documents, what topics emerged in each country, how those topics were compared and whether an LLM could keep pace. Every one of those rested on a choice we made.
This section shows what happens when those choices are made differently. It is not about whether the model was right or wrong. It is about the choices around it, how they cascade, and what they do to the result.
Remove a source group, keep the model and analytical rules fixed, and compare the result with an equal-sized random removal. See which findings survive and which depend on who was publishing.
Run one pooled model instead of three separate ones to see what merges, what stays distinct and what is left unassigned.
See what held, what moved, and what evaluation has to cover if a finding is going to be acted on.