By viewing the test case input as a string and sorting them it became clear that it would be possible to determine the correct output from only the first 2-4 characters of the string. This is optimal as it achieves a working (passes test cases) calculator with minimum effort even though it fails for edge conditions (anything other then a test case). Also the users interface is highly unfriendly - grouping the numbers and operators into arbitrary categories that presumably have no relevancy to the user.