Plain language that survives the compliance read
Science communication is translation with liability attached: say less than the paper found and it is dull; say more and it is a claim somebody will quote. The craft is calibrating strength while staying readable.
Why the mainstream APIs fail here
General models either overclaim, because confident copy reads better, or bury the finding in hedging after a safety pass. Both fail the compliance read, and neither keeps the mechanism honest when the topic is a pathogen, a carcinogen or a drug.
How MagmaRouter is used for this
Feed the paper and ask for the finding in one sentence, then the paragraph, then the page, at the paper's confidence level, with the limitation stated once and plainly. Ask where the draft overclaims; fix those first. Nothing is retained, which is what your legal reviewer will ask.
A working loop
- One sentence per finding, calibrated to the paper, before any copy.
- State the limitation once, plainly; repeating it reads as doubt.
- Have the model attack its own draft for overclaims; then fix them.
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.