The correct answer is C because brand voice and company-specific quality standards are custom evaluation criteria. A standard coherence check can confirm whether a response is understandable, but it cannot fully judge whether the response matches a company’s unique tone, phrasing, and service standards. A custom evaluation using an LLM judge lets the team define explicit criteria, such as “empathetic,” “premium brand tone,” “no unsupported promises,” or “uses approved terminology,” and then score responses against those rules. Option A is too narrow because coherence is only one generic quality dimension. Option B is too generic for brand-specific scoring. Salesforce Testing Center documentation explains that an LLM judge can compare an agent response against evaluation criteria such as factual accuracy, relevance, coherence, and faithfulness to source.