As collaborative learning is actualized through evolving dialogues, temporality inevitably matters for the analysis of collaborative learning. This study attempts to uncover sequential patterns that distinguish "productive" threads of knowledge-building discourse. A database of Grade 1-6 knowledge-building discourse was first coded for the posts' "contribution types" and discussion threads' "productivity." Two distinctive temporal analysis techniques--Lag-sequential Analysis (LsA) and Frequent Sequence Mining (FSM)--were subsequently applied to detecting sequential patterns among contribution types that distinguish productive threads. The findings of LsA indicated that threads that were characterized by mere opinion-giving did not achieve much progress, while threads having more transitions among questioning, obtaining information, working with information, and theorizing were more productive. FSM further uncovered from productive threads distinguishing frequent sequences involving sustained theorizing, integrated use of evidence, and problematization of proposed theories. Based on the significance of studying temporality in collaborative learning revealed in the study, we advocate for more analytics tapping into temporality of learning.