A contract-analysis assistant on the mid tier keeps giving wrong termination dates. Tracing shows retrieval returns only the first three chunks of each contract, while amendments that change the termination clause usually appear near the end. The team lead proposes moving to the largest tier because the reasoning is complex. What is the best next step?

Answer: A — Fix retrieval so amendments are returned with the clauses they modify, then rerun the eval before any model change

(a) is right: the model never sees the amendments, so no amount of capability or reasoning can use them; fix the context first and let the eval decide whether a model change is still needed. (b) and (c) buy capability for a context failure, adding cost without supplying the missing facts. (d) teaches the calculation, but examples can't make up for amendments that were never retrieved.