Hippocampal circuits form the basis of navigational cognitive maps. Yet spatial information alone is not enough to build a useful map, context cues need to be embedded in it. While the encoding of spatial information in the hippocampus is well-established and place cells, border cells and others have been identified, the precise representational strategies of valence information remain elusive. Here, we study how learning combines spatial and affective knowledge in the population activity of ventral-to-dorsal (valence-tuned) and dorsal-to-dorsal (space-tuned) pathways in the CA3 region. Do incoming axonal fibers carry exclusively spatial information that is invariant to the presence of affective information? Or is spatial information encoded along certain dimensions of axonal population activity, while complementary or the same subspaces within the same axonal population activity provide affective information? We provide evidence for the second hypothesis and show that both intermediate-to-dorsal and dorsal-to-dorsal CA3 axons form an invariant space-coding manifold that remains robust across tasks. Within this manifold, specific dimensions are present that inform about presence or absence of context cues without altering the spatial map. This suggests that the representational space in CA3 mixes spatial and affective cues but is simultaneously untangle them by considering different dimensions in the geometry of the population level activity.