Sample average approximation - Computing Upper Bound Estimates

There seem to be 2 versions of the Sample Average Approximation
algorithm (not original) - 1) From Santoso et. al (2005) and 2)
Linderoth et al. (2005). I seem to have some confusion about computing
upper bound estimates based on these two papers.

1) Santoso et al.(2005) compute an upper bound estimate by using just
one first stage solution (the solution corresponding to the min
objective value of the SAA problems). That is they use one first stage
solution, sample and solve multiple second stage problems, average the
second stage problem values and report the total of second and first
stage costs.

2) Linderoth et al.(2005) compute an upper bound estimate by using
each first stage solution obtained from each of the SAA problems. That
is FR EACH SAA solution, they sample, solve multiple second stage
problem, compute total objective and report the average of the UB
estimate for each solution.

Which is right?

Assuming 1) is right, we clearly have one first stage solution, which
can be used in practice.
Assuming 2) is right, should we use the first stage solution that gave
the lowest SAA objective?

In my implementation of the algorithm 1), I have some negative gaps,
which seemed incorrect (assuming my algorithm has been implemented
correctly), so I am having some doubts regarding 1).

Any insight is appreciated.

Thanks,
J