In regard to typos in hash keys, i have used various tied hashes that require registering any key before it is used. The definition of 'used' varies from implementation to implementation. In some cases i'd let it ask for existence of a non-registered key, but in other implementations, any action on a non-registered key is fatal. Etc, Etc, Etc.

My use of this approach is pretty rare, and is not perfect because its run-time checks, but in certain critical apps (such as bulk changes to my data warehouse) this approach can be advantageous.

I guess it comes down to a trade off of ease of use vs. level of checking.

---