NLSR

Experimentally, we should write the tests for the revised TLV system first, to see if it helps in our workflow.

Ideally, these tests would show something like, for class X:

• The function that serializes a given instance of X produces a string matching some expectation.
• The function that de-serializes a string representing an instance of X will produce an identical instance of X when de-serialized.
• Any separately-serializable components Y of X also fulfill the above two criteria.

As a corollary of the above, it should be possible to "fabricate" an instance of a given class, by hand.