Meridian wants the charter-desk assistant's answers to always end with a standard disclaimer sentence, word-for-word identical every time, regardless of what the retrieved clause says. Where is the most reliable place to enforce this exact, invariant text?

Programmatically appending the fixed disclaimer text after the model finishes generating is the most reliable approach because it removes the disclaimer from the model's generative process entirely - a deterministic string concatenation step guarantees word-for-word identical text every time, with zero risk of paraphrasing drift. Instructing the model to 'remember to add a disclaimer' without specifying exact wording leaves the precise phrasing up to the model, which will vary it across calls even with a low temperature, failing the 'word-for-word identical' requirement. Relying on the model to recall the correct wording from training data is unreliable and inappropriate - the model was never trained on Meridian's specific disclaimer text, and even if it approximated the intended phrasing, an approximation is not the exact invariant text required. Retrieving the disclaimer from the clause index conflates two different kinds of content - the disclaimer is a fixed piece of legal boilerplate unrelated to any specific contract clause, and routing it through semantic retrieval introduces unnecessary risk of it being mismatched, paraphrased, or omitted depending on what the retriever ranks as relevant. This item tests recognizing when a chain component should be deterministic code rather than a generative step at all.