Hallucination Risk Checklist

Model may fill gaps with plausible but unverified facts.

Recency claims are error-prone without explicit cutoff dates.

High-stakes topics require stronger evidence and traceability.

Forcing certainty suppresses honest uncertainty handling.

Exact numbers often require robust datasets and references.

Large-scope tasks with short context increase unsupported synthesis.

Without citation rules, unsupported claims are harder to detect.

Ambiguous references can cause incorrect entity resolution.

Conflicts increase behavior drift and inconsistent outputs.