A science club taught new members how to evaluate claims in short reports. After each lesson, students corrected their answers while the explanation remained on the screen. Almost everyone answered the same questions correctly on the second attempt. The club initially treated this as proof that the lesson had worked. The next day, however, students received a different report that used the same reasoning pattern but different vocabulary. Some repeated the original mistake. The club changed its review process. Students still corrected the first answer immediately, but they also wrote the rule they had used and returned a day later to a new example with no explanation visible. An answer was counted as stable only when the student could identify the relevant evidence and reject an exaggerated claim in the new report. Immediate correction remained useful, but it was treated as the beginning of review rather than its final test.