The parallel non-deterministic derivation mode ensures that a single node is deleted by at most one actual rule application. If several rule applications delete the same node, one actual application is chosen pseudo-randomly.

Assuming to apply a rule set:

 A ==> B A
 A ==> C A
 A ==> A A

on an axiom A

could result after two steps in:

• BBA
• CCA
• AAA
• BAA
• CBA
• …

Because at any step the software pics randomly one of the possible rules for A and applies it.