I think it's likely to be 0-2: just de-risking in a program still characterized as testing; or at least I understand it to still be thought of as a test program.

I can absolutely see not wanting to change too many variables while they're working out the full, targeted flight profile. I can also see, with sufficient confidence, that they'd make the line between test/production a bit fuzzy with carrying production payloads rather than mass simulators. Trade-offs... not much else to say.

For Starlink v3, I could absolutely see them delaying production starts until the capability to actually deliver them was closer to reality without regard to backlogs and schedule delays: why tie up capital in satellites sitting in a warehouse? Especially if you're also improving designs, adjusting those designs for new realities/tech/etc. getting ahead of that game would (my gut feeling) have more costs than benefits.