Unlike inter-hypothesis, running intra-hypothesis experiments allows comparing models ceteris paribus

When we run two experiments sharing identical infrastructure (intra-hypothesis), we can dial a single variable allowing us to check exactly how that change impacted the model performance. When running comparison between two architectures (inter-hypothesis), we’re unable to isolate a single change, since ceteris paribus (all else equal) does not apply here - the architectures are inherently different.