Monitoring one's learning activities is integral to self-regulated learning (SRL) and contributes significantly to successful learning outcomes. Judgments of learning (JOL), a crucial component of SRL, involve metacognitive assessments where individuals gauge their ability to recall learned material on future tests. While prior research underscores the link between accurate JOL and enhanced performance, there is a paucity of literature that studies JOL in resource-constrained countries and its temporal nature. Our study touches upon this gap by investigating the relationship between JOL and mathematics task performance while also examining the evolution of JOL over time among middle school learners from Kenya. Leveraging data from 317 students, our findings reveal that a majority of learners exhibit mostly accurate JOL (64.87%), with notable proportions being overestimated (31.33%) and a few students underestimated (3.80%). Moreover, learners initially demonstrating accurate JOL (62.73%) predominantly transition to correct JOL (60.39%), with a significant subset overestimating (31.68%) and a small subset underestimating (2.87%) their performance. Our research sheds light on the JOL variations in a different demography and highlights the stability in the temporal nature of JOL thus enriching our understanding of metacognitive processes, and informing the development of targeted interventions to enhance learning outcomes worldwide. [For the full proceedings, see ED665357.]