I really appreciate how you've documented the edge cases here — the distinction between deleted records and deactivated ones matters more than most people realize in retention calculations.

Confused by the 2007 value — is that a methodology change?

Exactly.