That rate seems much too high, for either a mechanical lock mechanism or an electronic keypad (which I guess would have some kind of lockout for brute forcing anyway).
I suspect that number is too high. 90 numbers per dial doesn't necessarily mean 90 valid positions. It could easily be 20 or 30 valid positions which comes out to a lot less. Plus, not all combinations of individually valid positions are themselves necessarily valid.
For example a standard Master dial lock has 40 hash marks but, IIRC, older ones had only about 12 valid positions. There was a trick you could use to find the third position, which would have left 144 combinations try except there were a few invalid combinations reducing to around 120. This is for a lock that a naive estimate would but at 40^3 or 64,000 combinations.
Master Locks have always had little hacks to them - for many years you could find the first position by feeling the tension difference, then the second and third positions were at predefined offsets which would change every few years as the offsets became widely known (pre-internet). My favorites were the keypad locks that allowed the numbers to be entered in any order - after a few weeks of use the paint would wear off the numbers being used so you didn’t need to guess at all - just be patient. Fast forward to today and you have two dozen guys on YouTube tearing down the locks, and you realize that most are only good enough to keep honest people honest.
They may not know how much the slop is, or the exact procedure to open it if you do know the combination (turn left X times, then turn right Y times...). Though you'd expect a locksmith with access to a safe-cracking robot would also have access to documentation showing that.
Or as others said, the lock mechanism could be faulty or seized.
The lock might be broken? Gummed up lubricants come to mind.
No idea if that's a common problem with safes, but it's pretty clear that a safe would not have any outside maintenance access to fix a problem like that (other than drilling it open, which seems to be routine enough given the "yeah, we can make it usable again" offer)
Depending on the quality of the lock there is some play in the mechanism (that is also why lockpicking works). This reduces the number of combinations. If you can lookup (model type documentation) or learn (by feel) how the internal mechanism works you can reduce the amount of combinations to guess even more as you can approach the combination one 'layer' at a time.
I once witnessed someone open a safe in a few hours. It is truly amazing to see these kind of proffesionals at work. How they approach this puzzle and the knowledge they can acquire by just feeling. Especially considering it took me a few tries to open the safe when I knew the combination.
But its not likely to tell the difference between 50-0-50 and 49-0-49. I suspect the "70% chance" for the robot to succeed is because it's going to try every 4th number or 5th number to hugely reduce the search space.
Trying every 5th number is just 8000 combinations!
