why order of atoms in body part of pattern affects results, best practices

[kliegr]

The same task with swapped atom patterns produces a vastly different result:

 ?b <molecularly_interacts_with> ?a & ?b <molecularly_interacts_with> <P12821> => ?a interacts_with <Q9BYF1>

results in an empty set after 2 seconds. When atoms in the body are swapped

 ?b <molecularly_interacts_with> <P12821> & ?b <molecularly_interacts_with> ?a &  => ?a interacts_with <Q9BYF1>

the result is 10000 rules after 12 mins.

(in both cases, "orderless" was disabled)

What is the recommended way of ordering atom patterns?

When the orderless option is enabled, the mining takes much longer. After 30 minutes, the progress is

  Action Amie rules mining, stage 2 of 4, steps: 8818 of 43157, progress: 20.0% (1.45 per sec ) -- processed rules, found closed rules: 0,

task_order.zip