UC-05 · Group B — Level 1 · Semantic Governance
The agent guesses the meaning — plausibly, and wrong
QuantData asks which reading was meant instead of running the wrong one — meaning is governed centrally, which is what semantic governance means.
The agent compiles a best-effort proposed SemanticTarget for “Hedge all our accounts back to USD.” and QuantData answers with the candidate readings of accounts side by side — bank, GL, client, trading, legal-entity — each with the trade it would produce.
The agent is allowed to guess; QuantData refuses to let a plausible-but-wrong guess execute. Meaning is centralized, not pushed onto the AI.
A real submit-operation (canonical_concept_ref finance.revenue, declared_downstream_use launch_planning) returns DecisionSummaryStatus: NEEDS_RESOLUTION with a blocking ResolutionItem of kind SEMANTIC_CLARIFICATION carrying candidate interpretation options; the required next call — submitResolutionAnswer — resolves it live to PROCEED.
“The agent guessed. It guessed plausibly. It guessed wrong. And it had no way to know.”