What stood out to me with SIGN was how one failed row can poison the next payout run.
Not the whole batch. One row.
Delegated execution already fired. Later that same row rejects. Support still sees the wallet as pending. Finance blocks the rerun until the operator can prove whether that row already cleared a settled step. Guess wrong here and the team can turn cleanup into a duplicate payout.
That is the real mess. One row, one wallet, two conflicting views, and one dangerous decision: does this row need a retry, or does it need to be left alone?
This is where SIGN starts to matter to me. The row-level state transition is there. The correction or supersession record is there. The safe-to-retry check does not have to be rebuilt from exports, screenshots, and memory right when the pressure is highest.
That still depends on teams modeling transitions well. Bad logic can still be preserved cleanly.
But this is why I'm watching $SIGN.
The hard part is not sending the batch.
It is proving one failed row is safe to resume, not replay.