It was 3 layers
- API interface (unpacking objects, etc)
- Business Layer
- DB/System Interface.
The business logic changes could often be totally contained, without updating the API or the database. It would be so obvious what layer needed to be edited, and the problems could be fixed so much faster. Rarely did a commit change two layers at once.
You can easily handle AuthN in the API layer, AuthZ in the business layer, and then access to a db in the system layer. Assume API is un-authZ’ed, assume the system layer is fully AuthZ’ed, etc.
There's a tradeoff between understanding the function itself and understanding how/where the function is used (and therefore understanding what will break when that function is changed).
The best heuristic I've got at the moment is roughly the square root of the size of the module. So, for the 10,000 line module, my vague instinct is that I'll have one hundred functions approximately one hundred lines each. That's not a hard and fast rule, just an observation about tradeoffs and I would definitely expect (quite possibly even most) functions to differ significantly without losing sleep over it.