User Interest Exploration

Core rule. Past behavior records what the system exposed and what the user did under that exposure; it is not a complete list of possible interests. An unfamiliar click is also not proof of a new preference. Read the page as an evidence chain: define the outcome, propose a direction, let the user steer it, help form a preference when knowledge is missing, learn across turns with a stopping rule, and test whether the change survives retraining and repeated exposure.

Evidence threshold. Treat every inferred new interest as a hypothesis. Stronger evidence comes from an informed user choice, successful correction, repeated or later behavior, or a causal comparison that controls exposure. If a result reports only list diversity, one click, or immediate satisfaction, call it broader exposure rather than lasting interest formation.