Intra-organisational inventor networks are continuously reshaped by project assignments, team restructuring, and geographic expansion, yet most research treats network structure as static and examines closure and brokerage in isolation from the spatial contexts in which they operate. This study asks how year-to-year changes in closure and brokerage within intra-organisational inventor networks influence firm-level innovative output, and how spatial dispersion moderates these relationships. Drawing on a 50-year panel of urban mining and WEEE recycling patents (1970 to 2020) spanning 572 firms, fixed-effects Poisson model estimated via quasi-maximum likelihood are estimated to examine the directional effects of structural change on patent counts. Changes in closure exhibit a robust inverted-U effect on innovative output, with the turning point falling within the observed sample range, and this effect is largely independent of geographic context. Changes in brokerage yield monotone positive returns whose magnitude is amplified by geographic dispersion in the primary FE-Poisson model specification, though this moderation is sensitive to model choice, while the amplification disappears for co-located firms, yielding flat associations. These findings demonstrate that closure and brokerage are temporally and spatially asymmetric mechanisms that must be modelled and managed independently. Together, they reframe the R&D network design problem from one of optimal structure to one of optimal structural trajectory.