base64 content encodings are unusable in security conscious organizations

Yes, integral base64 encodings for cell content is nice from network, efficiency, and file management aspects.

Unfortunately, security conscious organization that have established review processes simply will not allow base64 content -- it is too easy to hide content or malware that way from their established (read unchangeable) processes.

For example, if an ipynb file has a matplotlib produced png image, such an organization needs that image referenced by the .ipynb but stored in an external png file. This allows existing processes for approved file types to operate on the cell content/results.

No files, tests, or entry points are named in the issue. Start by reviewing how nbformat represents cell outputs and external media; done would require a clearly defined way for PNG output to be referenced outside the .ipynb instead of being embedded as base64.

Written by the indexing model from the issue text.