Yes, Alice would get back the full amount (Collateral+Payment) upon expiry...and Bob obviously loses the real goods (to Alice). Those are the respective Phase 2 gains and losses which I highlight in my original post.

Bear in mind that the algorithm that I'm proposing cannot eliminate all the risks and unfortunately for Bob, he bears the ultimate risk. But the point of a (potentially very large) collateral and/or a (potentially very long) expiry time is to incentivise Alice to hold up her side of the exchange to get her collateral back in a timely fashion.

If Bob is undertaking several bitcoin-based exchanges with different people then it's possible that he might incorporate a risk premium into the Payment amount for each exchange (as compared to a smaller amount of fiat currency that he would otherwise accept in an equivalent series of real world exchanges for the same goods) reckoning on still being able to make a profit on the whole set of exchanges even if a proportion of the exchanges go bad.