Visual features are spatiotemporally integrated along motion trajectories. For instance, when a central line is followed by pairs of flanking lines, two motion streams diverging from the center are perceived. The central line is rendered unconscious by the subsequent flanking lines through metacontrast masking. Surprisingly, if the invisible central line is offset, the entire stream appears offset, even though the flanking lines are, in fact, straight. Further, if one of the flanking lines is offset in the other direction, the two offsets integrate and cancel each other out. This integration is mandatory and occurs only when the offsets are presented within a specific, temporal window, starting with stimulus onset and lasting for about 450ms. Here, we asked what determines the extent of this unconscious integration. Observers discriminated the perceived offset of the motion streams. In line with most models of decision making, one might expect that the window terminates as soon as sufficient evidence about the offset is accumulated. However, this is not what we found. We presented either a large offset at the first line or smaller offsets at the following lines, all in the same direction. When performance was the same in both conditions, the duration of the integration windows was identical. Hence, it does not matter whether strong evidence is presented right from the beginning or dispersed along the motion stream. When we increased the processing load by adding two further offsets at two additional lines, which, however, canceled each other out, the window duration increased slightly. Lastly, we found that absolute time determines the window duration but that the number of lines or the ISI between the lines do not. We propose that perception is a series of discrete frames, which depends mainly on absolute time, potentially on the processing load, but not on stimulus evidence.