> Unlike code, natural language laws have many exceptions and context dependencies,

In the interpretation of the law, yes, but that doesn't matter for version controlling the text of the law, and in that sense, you can absolutely manage the text pretty cleanly in git. Most laws are just text. Many bills are literal git-diffs, except in prose, which can make it hard to understand a bill.

> exception handling (hardcode-style responses)

This is basically non-sequitur (there no such thing as a "hardcode-style response" in git…), and again, I think you're misunderstanding the idea here.

This brings us to my even more unlikely idea - when whoever wrote that line in the legislation dies or retires, the line enters a short review period where it's either renewed with a new author on the line to speak to it or is deprecated. This way someone would always be there to just answer for the intent. Also would help garbage collect legislation that is no longer useful.

> natural language laws have many exceptions and context dependencies

The complexity is in common law, not natural-language laws.

The U.S. Code and Code of Federal Regulations can be completely presented through a Git-like system. The complexity comes in when attaching court cases.