- make records of the smallest parts that are practical, and content address them
- make tables of these records that are essentially collections of content hashes
- this enables multi-client support natively
- how?
- we only need to create records that dont exist, we can reuse those that do
- when we derive data, we need to include the content hashes of the data we are deriving
- for multi-client
- this means that when two clients have different versions of the source, the areas that overlap, overlap deep into the compilation pipeline
-