UPDATE
All of the relationship graphs have been untangled. We've undertaken a complete teardown. We didn't just prettify the graph, we rewired it and ditched a lot of TOs. The new system is in testing with betas and we're getting great results. Big performance gains and only minor problems cropping up.
A big job I'll bet especially when it comes to ditching TO's. Did you use Base Elements or FM Perception ?
We run a custom process over the output of Copy as XML to produce plain text documents. The capture eall the major elements while ignoring the minutae (such as hex flags). Those files get pushed into an SVN repository and we pull an update to get a copy onto our local machines. If we need to go back in time, it's all there in the SVN going back a long way, so we can run a diff against any point in the past.
Using BBEdit or TextMate we can easily search across the entire project for any object. So it's fairly easy to determine exactly what will break when you touch something. I also use BaseX as general XML browser and search engine for more exotic queries.
Once you have visibility it is much easier to identify relationships that were unnecessarily duplicated, scripts that were not referenced anywhere, etc.
I posted the other day about Notenik. It has became a real productivity tool for me. My hour-by-hour workflow at the moment looks like this:
,-----------------------------------------------------,
| |
| .--> SVN --> BBEdit --> Notenik --> FileMaker--. |
| | | |
| '-----------------------------------------------' |
| |
'-----------------------------------------------------'