The part in SIGN that felt most real to me was one damaged row after a payout run broke halfway through.
One wallet already got paid. The next row failed. The row after that belonged to a valid claimant, showed as completed for a moment, then the batch died before the final settlement state was safely recorded.
Now the operator is trapped on one decision. Resume from that row, skip past it, or restart earlier.
Resume too early and that same claimant can get paid twice. Resume too late and a valid payout disappears from the run. The mistake is no longer in the original table. It is in the recovery point.
That is where SIGN feels sharp to me. TokenTable matters here because the operator needs row-level execution evidence, replay-able row history, versioned state, and proof of whether that row actually crossed settlement or only looked completed before the process died.
If the system cannot prove that one row, retry becomes guesswork.
That is the question I keep coming back to in SIGN: after one broken batch, can the operator resume from the exact damaged row without duplicating one claimant or skipping them entirely?