the real lesson to be learned here is that ultimately scenes are not to be trusted, so try and rely on them as little as possible . not an easy task, but there are some good habits that can go a long way in making this work for you.
models (emdl) are your friend. a major part what you produce can be put in one - corruption in models exists too, but is less frequent in my experience. I usually keep a history of scenes in which I edit the models (so with the model local to the scene) - and export milestones to a model, occasionally keeping backups of the model. with two parallel systems to save data, each with versioning/backups if you choose, it is a lot less likely that you’ll loose a lot of work when a scene (or model) corrupts. from the scene you can produce the model again, and the other way around within limits. ‘big complex scenes’ or ‘production scenes’ should ideally be just a collection of referenced models – perhaps created by scripting even? as in: import all ref models to build the scene, set them up for the shot, apply animation data, set render settings, build passes, put models in partitions. Standardize on a system for this, with proper naming, and so much can be automated – even with very limited scripting skills. There’s other types of data that can be used too – presets of all kinds, compounds, EANI, pointcaches, .xsi and .fbx files. If you wanna be real smart about it XML files. Lose a scene? Never mind, you have still have the bits the scene was made of, and can recreate it. I know its not as trivial as I make it sound, but ’not to trust scenes’ was one of the best lessons I learned. From: Octavian Ureche Sent: Thursday, July 05, 2012 3:56 PM To: softimage@listproc.autodesk.com Subject: Re: WHATTA HELL??????? So why would anyone care about xsi having ascii scene file description? Well....this is why. I feel your pain, though trying to recover a scene in this situation, has more to do with luck than anything else.