The
retention_start_date field was added in 2019, retroactively populated for prior years using a different method than the native calculation. This creates a subtle incomparability: 2019-onward retention is calculated one way, while earlier years were backfilled using a different approach. Should we recalculate the entire series using the current method? (edited to fix a unit)The 2009 break in retention
Confused by the 2002 value — is that a definition change?