Interesting write-up! One concern: committing to a merkle root with different sighashes in the scriptPubKey seems tricky, since you’d need the inputs to compute the sighash—but the inputs depend on the finalized scriptPubKey. The classic circular dependency issue.
I’m assuming everything on the stack is from the witness stack, so I can see some interesting possibilities with a MuSig2 <key> and presigned transactions. I’ll go back and read your bidding problem post—maybe that’s the intended use case. If so, even combining the ‘Schnorr trick’ with MuSig2 keys could lead to some cool results.