Failure mode
Intent Mismatch
A failure mode where the system answers a plausible question, but not the one the user or downstream agent actually meant.
Domain: semantic-reliability
ARF layer: semantic-integrityARF layer: ai-readiness-interoperability
Signals
- ambiguous entity resolution or poor metric naming leads the system to choose the wrong object
- natural-language prompts produce confident, grounded, but irrelevant answers
- retrieval is technically correct but semantically mis-scoped
Why it matters
This is one of the most important failure modes in conversational and agentic systems. The model may not be hallucinating. It may simply be resolving the wrong intent because the platform lacks the semantic context to constrain the question correctly.
Prevention patterns
- stronger identity and semantic contracts
- canonical naming plus alias handling
- graph- or relationship-aware context for entity resolution