Relevant bit:
This sounds good but we’re not into it. I mean, first, we do have plenty of positive reinforcement in the form of pretty graphs and the satisfaction of adding datapoints. You can even spin the pledges as positive — they help you quantify the value of your goals. That can be powerful information for us rationality nerds.
But why not reframe Beeminder to focus on rewards? Well, paying money up front and getting it back unless you derail is a trick — it’s equivalent to getting stung. At least for me personally, the equivalency would always be at the back of my mind and bother me.
And there are more pragmatic problems. I like having scary high pledges on some of my goals. It would feel especially unreasonable to pay up front on those. Even more pragmatically, most goals are open-ended: get 10k steps (or work 40 hours, or practice piano for half an hour or whatever) per day forever. There’s typically no particular point when it makes sense to get your money back. It would be totally inefficient to have money always flowing back and forth and would really muddy the mental accounting in terms of how much you’re paying Beeminder for the motivation it’s giving you.
Not to mention the laws and accounting involved. We’d be kind of a bank and have revenue that wouldn’t count as revenue. I assume this part would be perfectly overcomeable if we were convinced the psychology / behavioral economics were right. But, again, we are not.
Setting money aside in escrow is definitely an interesting idea, but as you say, requires quite a lot of trust. I did toy with idea of rewarding people with the punishment money from other failed commitments but it'd be quite tricky to make the rewards worthwhile, while making sure the system doesn't get gamed.
Very short version: We don't think smart contracts add much. Your users trust you.
For some definition of "motivating", probably so, but I wouldn't want to take that extra stress to keep a self-motivated minor commitment.