This paper re-validates middle school students' learning progression of the concept of buoyancy (Gao et al., 2020) by using two-time point data. The previous study developed a four-level learning progression based on data collected at a single time point using cross-sectional design. This single-time-point model of learning progression calls for a further validation through a longitudinal design focusing on changes in individual students as a result of targeted instruction. In this study, 91 eighth-grade students first learned buoyancy under a traditional pedagogy and took test "A" of buoyancy. Based on results of test "A," the teacher provided a targeted remediation instruction to students. Students then took a second test "B" which was designed as a parallel instrument to test "A." Latent class analysis was conducted to locate students' levels in the learning progression, and a rule space model was employed to diagnose students' concept mastery patterns at each level. The test "A" data provide evidence to support the original learning progression of buoyancy developed by Gao et al. (2020), but the test "B" data suggest two more additional progression sub-levels for the prior study. Together, our findings indicate that one-time point data under the current status quo of teaching might be insufficient to accurately depict pathways of scientific thinking for how students progress from the lower anchor to the upper anchor. We thus argue that it is essential to involve well-designed instruction and collect multi-time-point data that could portray alternative pathways in students' learning progression.