When OPA builds the rulebase, it takes all of the compiled rules and sorts them into an optimised order.
It does this so that all of the rules can be executed in a highly efficient linear left-to-right sweep. This is all explained in the following Oracle Whitepaper - "Linear Inferencing - High Performance Processing". Page 2 is the most relevant:
So... you could predict which attributes will be set next when base attribute(s) have been changed. You can do this by looking at the structure of the rulebase and seeing which inferred attributes (via rules) are set when the base attribute is changed.
Do a visual browser on a base attribute and "generate influenced rules"... effectively by doing this you are forward-chaining and that is what the Engine will do once a base attribute is changed.
Try this out in the debugger. Set a base attribute to a different value, and you will see the forward chaining impact ... all of the changed attributes from this cycle will be highlighted in a green background in the debugger...You could check this matches your "prediction" from the visual browser file...
Hope this helps... let me know if this is not clear.