Purpose: compress rules into model weights; prompt = brief summary.

Behavior: robust slot-filling (budget, district, timing), clear next steps.

Data: harder dialogs; multi-constraint matching and scheduling.

Context: short control prompt; query-centric conditioning.

Output style: structured, action-oriented, minimal boilerplate.

Intended use: near query-only inference with small header.

Limits: may under-specify rare policy clauses without prompt aid.

Safety: aligned; escalates unclear/legal topics.