Note also that the nominal starter quality varies within the team as well as among teams.

Then consider that most consider the nominal starter, the starter of record on the opening game of the season. If the real starter is already injured then the play is of the unit is begins from an already impacted state, though the stats will not reflect that situation.

That is why I said that FO, flawed as they are, provides the best data available that I am aware of.

That’s a point I hadn’t really considered. That’s a hard thing to work in to a metric because the importance of continuity varies greatly between units. Obviously continuity on the offensive line is very important, but I would think it’s not as important at a position like wide receiver or defensive tackle, where player roles tend to be fairly individual and similar between different schemes.