scheduling

There is not a graphing switch like there is for "sort by data spread" that would sort in the graph by start time.

What you need is a criteria that can vary only by activity, and does not depend on anything else -- i.e., if you add any indexes to any of the durations. The challenge with that is when you do what-if, you might get different starting time orderings in different scenarios. This happens, of course, in the Monte Carlo runs when your durations are also uncertain. 

There are several possibilities for an invariant ordering, but it might not be ordered for specific scenarios where the start times are in different orders. For example, you could take the minimum starting time that occurs anywhere in the mid-value.

Once you have settled on a simple criteria to sort on, then the most obvious thing to do would be to add a new index, Sorted_activity, defined using SortIndex(...) on the simple criteria. You would then have a different node for Earliest_start_time that is indexed by Sorted_activity.

A next question to ask is whether it could be done without adding any new nodes. On the surface, it seems impossible because the sorted_activity index needs to use Earliest_start_time, but Earliest_start_time needs the list of activities to compute (the unsorted list). But there is a possible trick for this. What you can do is define:

Index Activity_idx ::= ComputedBy( Earliest_start_time )

Objective Earliest_startTime ::=
   LocalIndex A := #SetDifference( \(outputs of Activity), HandleFromIdentifier(identifier of self));
   Local start_time_mid := Mid(Activity_start(A));
   Local inds := IndexesOf(start_time_mid, except:A);
   Local sortCrit := if size(inds)=0 then start_time_mid Else Min( start_time_mid, ...inds);
   Activity_idx := sortIndex(sortCrit);
   If IsSampleEvalMode 
	Then Activity_start(Activity_idx)
	Else start_time_mid[ A = Activity_idx ];

The trick part is the use of ComputedBy by the index, so that the start times can be computed in a local variable before you commit to the final sort order. I've add in the other things discussed above -- i.e., a sort criteria that uses the min of the mid-value start time along all other additional indexes that might appear. The definition of the local index A is just the former definition of Activity_idx.