The first explanation must not choose all the evidence
After a striking pattern is found, it is easy to search only for confirmation. Ariadna should expose that risk. A useful inquiry also asks which document, date or relationship would undermine the initial suspicion and which part of the context has not yet been examined.
What the system contributes
Architecture prevents a model from deciding by itself which evidence is allowed to exist. A rules-based baseline expansion runs before model prioritization; hypotheses must search for confirming evidence, evidence capable of destroying them and observations that discriminate between rival explanations. Uninvestigated eligible links remain recorded.
Look for legitimate explanations too
Companies may share professional advisers for ordinary administrative reasons. A change in officers may reflect a documented reorganisation. These possibilities are not obstacles to investigation; they are alternatives to test. The work becomes stronger when a hypothesis can lose support in the face of contrary evidence.
Record the boundaries of the search
The team needs to know which registers were consulted, which failed to respond and which paths remain unexplored. That helps direct the next review. The purpose is not to exhaust the internet or call an inquiry complete because it returned many results, but to preserve explainable coverage and open questions.