The error amplifier will not always have a negative output and not always have a positive output. This is the point of the error amplifier.
Because of the negative-feedback characteristic of the op-amp it will try to keep 2.5V at its negative pin, as you say yourself. So if the incoming signal over the potentiometer P1 drops below that it will swing up the output to start trying to pull it up. When the signal incoming goes above the target 2.5V it will swing the output low.
There is your feedback, that signal gets propagated through the current sense amplifier, which is yet another kind of error amplifier, but for the primary current, as a protection, into the control circuitry.
The P1 potentiometer allows you to adjust "how violently" it reacts to errors in the output. The lower its value the more heavily the error amplifier will react to small errors on the red arrow.
The point of all the capacitance in the feedback, is that it is not at all a DC signal. The switching regulator switches the output on and off with a frequency, likely between 20kHz and 200kHz (I have not looked up this exact chip, but those are common fly-back outer frequencies). So, the capacitances in the feedback loop allow the error amplifier to filter out frequency responses that are too low or too high, by adjusting the feedback impedance depending on the frequency it tries to amplify.