Campaign-season questions to the city agent must not become political advice. How should NeMo Guardrails dialog policy handle them?

During campaign season, questions to a city agent can slide into political advice or electioneering—activities municipal bots must avoid. NeMo Guardrails dialog policy should escalate political campaign questions to a human clerk script via guardrail flows. This preserves resident service continuity while keeping the automated agent out of endorsements, ballot-measure advocacy, and campaign strategy that would undermine neutrality and public trust. Answering with candidate endorsements generated from social media sentiment fails because cities must not endorse candidates, and social sentiment is not an authoritative civic source. Ignoring guardrails and letting the model debate ballot measures at length fails because extended political debate is the out-of-policy behavior rails exist to stop. Silently inventing neutral-sounding campaign strategy without disclosure fails because it is covert political assistance and still crosses the no-electioneering boundary. Exam caveat: not every election-adjacent question is campaign advice; factual pointers to official election office hours, registration deadlines, or published sample ballots may be in scope when drawn from authoritative sources. Classify intent so campaign strategy and endorsements escalate, while verified election-administration facts may be answered with citations. Practical check: test endorsement requests, “who should I vote for,” and strategy prompts for escalation to the clerk script; separately verify that factual election-office FAQs retrieve from the official corpus, and review rail logs for false positives that block neutral administrative answers.