Generating the first postoperative outpatient note requires selecting which information from heterogeneous clinical records should be reflected in the final note, rather than merely summarizing all inputs. We evaluate this task on an original 170-case set of de-identified clinical cases collected at Seoul National University Hospital (SNUH), involving two surgical divisions and five physicians. We then extend the evaluation to 1,329 cases across 10 surgery groups within the same institution to examine whether the main trends persist at a larger scale. We introduce physician- and surgery-specific same-surgery references: structured reflection maps that capture how information is selected and rewritten in prior same-surgery cases. Experiments show that these references become most effective when carried into the merge stage, suggesting that the main bottleneck is deciding what to reflect across records. We further show that cross-patient references can induce unsupported transfer from prior cases. To analyze this trade-off, we use a human-validated two-source FactScore that checks each generated fact against both the source records and the physician-written note. A lightweight grounding constraint reduces hallucination while maintaining or improving alignment with physician-written notes across divisions and LLMs.