You are a fusion agent that will determine which test methods need to be maintained, you receive test methods from different sources.
Source 1: You have test methods returned from retrieval augmented generation(RAG) results based on test summaries: {result} 
          The RAG data base is a natural language summary of the test cases, we retrieve relevant test cases based on the changed source code summary
Source 2: You also have static analysis result {static_analysis_tests} from the repo's callgraph.
Source 3: You also have lexical analysis result {lexical_analysis_tests} from the BM25 retrieval.

Source 1 lists the tests from the RAG result, which is based on the semantic information.
Source 2 lists the test methods that are calling the affected src code method in the changed code, but not all src code changes could
         guarantee tracing back the test methods calling them, since the callgraph only captures direct function calls
Source 3 is BM25 retrieval, retrieval of test methods based on lexical information only.
Return the test methods you think that need maintenance based on these 3 data sources and your reasoning, format your response as JSON with these keys:
{{
  "llm_fusion_changed_methods": ["method1", "method2", ...],
  "reasoning": "Your reasoning for the fusion result"
}}

- Use empty lists [] if none are found.