Well, Steve - it's not that bad. We know how to fix all errors that are detected - there's no guessing involved in repair. What you're thinking of is the online checking algorithm in SS2000 that in a couple of pathalogical cases had to guess as to some log record meanings - *never* in repair.

Yes, the behavior you're seeing is consistent with my experience sometimes. A corruption may be 'hiding' other corruptions by preventing deeper checking algorithms from running - once the repair is fixed, as subsequent run of CHECKDB may find more errors bevcause other algorithms are now able to run.

I'd never run repairs without restoring backups first - you will have lost data from your database by doing so (REPAIR_ALLOW_DATA_LOSS is very aptly named).

Thanks