Interesting remark, thanks.
I’ve take a look about nSequence usage onchain with this [tool](nsequence_runner) and what I can observe is:
- there is no nsequence bit that have not been used at least once
- the combination of bit31 = 0 && bit 22 = 0 && bit 21 = 1 && (nSequence & 0xFFFF) > 0 && the spending script contains an OP_CSV (only looked at segwitv0 & taproot)
I’ve no strong opinion on how it should be interpreted btw