I have found when a child trigger is in a problem state along with the parent trigger in a problem state, the child trigger state it is not re-evaluated using latest data after action escalation continues from a paused state once the parent trigger recovers (even if the latest data suggests the child trigger should have recovered).

• Create trigger A
• Create trigger B dependent on trigger A
• Create action with an operation that starts after period of 300 seconds

(Next steps all must occur within the 300 seconds)

• Send data to ensure trigger B is in problem state
• Send data to Ensure trigger A is in problem state
• Send data to Ensure trigger B is in ok state
• Send data to Ensure trigger A is in ok state
• wait and see that the action operation for trigger B actually fires

What I believe should happen is when the data is sent to ensure trigger B is on ok state while trigger A is in a problem state that it is re-evaluated and the action cancelled.