Studying the propaganda means reading the propaganda.
The field exists to understand how this material persuades people. That work is impossible against a model that will not look at it.
Academic researchers, counter-extremism units, platform policy teams, and the NGOs that report to them.
Why the mainstream APIs fail here
A coded corpus of recruitment posts is the raw material of the discipline. Hosted models refuse to summarise it, classify it or explain its rhetorical moves, so the researcher hand-codes thousands of documents and the sample shrinks to whatever one person can read. The refusal does not reduce the harm; it reduces the study of it.
How MagmaRouter is used for this
Classify, summarise and cluster the corpus in a single pass, including the passages that make the refusal fire. Per-key caps keep a grant-funded project inside its budget, and the no-retention policy is what an ethics committee will want in writing.
A working loop
- Take the cap from the grant line and write it on the key.
- Run classification across the whole corpus rather than a safe subset of it.
- Record the receipt hashes in the data management plan as evidence of what left the institution.
Prompts are not stored. A signed receipt of the hash comes back on every call, so you can prove later what MagmaRouter logged. Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.