A goal seeking structure with delay can be responsible for behavior that's quite difficult to intuit. There is also a video for this insight with is a component of the Systems Archetypes Course.
The Goal Seeking with Delay structure is the the basic goal seeking structure with a delay added either between the [action] and the [current state] or between the [current state] and the realization the [current state] has changed at the [gap]. This means the next [action] isn't informed by the actual [current state] of things.
The difficulty this causes is that the [gap] that influences [action] is not based on the actual [current state] and as such presents the possibility of the [action] resulting in overshooting the [desired state].
This implementation of the structure has added [delay state] governed by [time delay] so the delay is easily changed for testing and an [action factor] to govern the percent of the [gap] that results in [action].
If we begin generic situation with [desired state] = 1; [current state] = 0; and [time delay] = 0 the results are those of the standard goal seeking structure.
If we introduce [time delay] = 1 you can see that the [current state] slightly overshoots the [desired state]. To correct for this the [gap] is negative and [action] changes the [current state] in the other direction. Notice the arrow for [action] has an arrow at each end.
With [time delay] = 2 you can see that the oscillations are even more pronounced though the model does finally stabilize.
With [time delay] = 3 you can see that the oscillations are even more pronounced and the model never does stabilize.
With [time delay] = 4 you can see that the model is very definitely out of control.
This is the one of the best examples of how the behavior of even simple structures are difficult to intuit from a qualitative diagram. At times only a simulation will provide the requisite sense of the implications of the relationships.
Strategy: Once you sense that you're dealing with this structure there are three possible strategies: 1) reduce or remove the delay; 2) delay the action until the current state is sensed; 3) reduce the action to make smaller adjustments.
Alter the parameters and Run Simulation to get a better feel for this model. Specifically what effect does changing [action factor] have on the behavior with the same [time delay]? Why? Also change the Simulation Time Step in Settings to 1, .5, .25, and .125 to see how this affects the nature of the graphs.