The schema-as-inference-input bit is the useful cut. Asking an LLM to emit JSON and then repairing it is still generation-with-validation; declaring the legal choices first means the model cannot invent "priority support" as a new team.
Cascade of rules → decision model → LLM → human matches how production already budgets cost. Calibrated probabilities beat reading confidence out of fluent prose.
melonape
Full-stack engineer, 10 years building for the web. Writing about AI tools, open source and the small tools I make.
The schema-as-inference-input bit is the useful cut. Asking an LLM to emit JSON and then repairing it is still generation-with-validation; declaring the legal choices first means the model cannot invent "priority support" as a new team.
Cascade of rules → decision model → LLM → human matches how production already budgets cost. Calibrated probabilities beat reading confidence out of fluent prose.