Research has shown that spanning a structural hole creates the advantages of information integration and control for the broker. However, scholars have observed that brokerage is a fragile relationship, implying greater instability of the brokerage structure than other network structures, as their benefits quickly disappear. Brokers play a key role in disrupting the natural tendency of structural roles and driving continuous changes in existing network patterns. They can choose either the tertius iungens or tertius eliminatus strategy to close the structural hole. However, while recent studies have begun to disentangle the mechanisms underlying the generation and dynamics of these structural holes, the scholarly understanding of when and why brokerage terminates through structural hole filling in these two ways is limited. Existing research has also ignored the role of knowledge content in structural hole-filling.Therefore, this study examines the relationship between the content of network knowledge and structural hole-filling evolutionary trajectories from the perspective of broker roles. We investigate the end of brokerage within an overarching framework by integrating the inducements and opportunities of both brokers and alters to fill or not fill the hole that separates them in the brokerage triad. Specifically, this paper takes 36,338 patents applied for by 41 large knowledge-intensive enterprises in China from 1996 to 2008 as research data. A two-stage Heckman model was constructed to correct for sample selection bias. We then use a multinomial logit model to test our hypotheses and discuss the impact of innovation output under different padding methods based on the citation information of patents in the sample as of 2020.The empirical results show that 1) when the knowledge bases of the broker and alters take the form of itinerant brokerage (H-H-L), they take the form of closure. 2) In the representative type of brokerage (L-H-H), the results were not significant for either disintermediation or closure. In innovation settings, disintermediation, which involves cutting out the broker, may only be legitimate when it provides clear benefits of effectiveness. 3) For the liaison triad (H-H-H), we find support for the hypothesis of closing by closure. 4) Regarding the closing of combiner brokerage (L-L-H), our results support closure; however, the result for closing by disintermediation was also significant. We spelled out two competing logics on the part of the broker and the alters and hypothesized that the combined effect would lead to closure. However, the motivation to connect directly to some of the alters is sufficiently strong to achieve disintermediation. In such cases, there may be a contingency that we were unable to identify when combiner brokerage results in disintermediation versus closure. Our brokerage typology results are largely supported except for the representative (LHH). The results of the nonparametric tests show no significant difference in the impact of innovation output under the different filling methods.In summary, this study contributes to knowledge of dynamic networks in several ways. First, based on the role of the broker, this study defines five different types of structural holes with the knowledge of the broker and alters as the boundary to study the structural hole-filling evolutionary trajectories. This also illustrates at the micro level that the analysis of network structure should use finer-grained classifications in subsequent research to facilitate more in-depth discussions of its evolutionary mechanism. Second, this study considers the perspective of all the actors in the brokerage triad and views the change in the network structure as a result of the comprehensive influences of a series of factors, which mainly come from the network environment embedded by multiple actors and the attributes of the actors themselves. This approach provides a more complete analytical framework for subsequent research on dynamic organizational networks and suggests a fundamentally pluralistic view of the drivers of network actions. Thus, we reinforce the foundational intuition that social structures are not simply the outcomes of individualized action or agency on the part of a single powerful actor, but rather evolve as a result of complex motivations and interactions among a plurality of the actors that form the social structure. Third, by considering content alongside the brokerage structure, we also escape the pitfall of confounding the forest for the trees, and vice versa. This research logic is important for the study of dynamic organizational networks, and relevant studies need to analyze the role of structure and its possible changes after controlling for the influence of network content. Finally, this study revealed the relationship between specific characteristics of the knowledge distribution among individuals in an organization and structural hole-filling. In fact, it establishes a theoretical connection between the knowledge-based view and structural hole-filling and expands the cognitive boundary of knowledge and structural hole-filling in existing research. Due to the uneven distribution of knowledge resources, there are differences in the knowledge possessed by individuals in the organization, which affects the change in network structure in different ways, especially the filling of structural holes.