Primer is an exceedingly complex and sophisticated movie. The high-quality film that director Shane Carruth used was by far the greatest investment in the $7000 budget, so almost every scene was shot precisely once and additional scenes providing character background and exposition, which, while not strictly necessary, would have gone a long way towards making the film more comprehensible, were never filmed, for the sake of expense. I do intend to create a writeup which explains the full sequence of events in the movie, but here, first, is some prerequisite knowledge: how time travel works.
The channel
The time travel boxes used in the movie work like this. When the box is switched on, a "channel" begins in spacetime. That is point A. Call it 12:00pm. Over the next few seconds or minutes (depending on the box size) an electromagnetic field builds up gradually in a parabolic form until it plateaus.
Later, the box is turned off. This causes the field to diminish parabolically to zero. This closes the channel at point B, let us say at 12:01pm.
A weeble placed within the confines of the channel - that is, within the boundaries of the box and between the times of 12:00pm and 12:01pm - does not move through time in the normal sense, because time does not flow within the channel in the same way that it does in reality. Instead, the weeble wanders from point A to point B and back again, backwards and forwards, over and over, in an indeterminate state, gradually accumulating time from its point of view.
This continues until the weeble's waveform collapses and it is removed - either at point A or point B. Every time the weeble reaches end A or end B of the channel, there is a finite probability that the waveform will collapse and the weeble will exit, as opposed to continuing on another trip along the channel. Actual hard figures about this are fairly sketchy in the movie, but it seems like the probability of exiting the channel is about one in ten thousand. After roughly 1300 trips from A to B, the probability is better than even that the weeble will exit. So, the amount of time that the weeble spends in the channel (from the weeble's point of view) is variable, but usually works out to roughly 1300 minutes.
Collapsing the waveform is the tricky bit. As in the famous Schroedinger's Cat experiment, the waveform is only collapsed when the box is opened. Of necessity, when the object involved is an inanimate weeble, the box can only be opened from the outside, by Aaron and Abe, at point B, at 12:01pm. Therefore, the weeble always exits at 12:01pm.
- If the weeble is put in at point A, the amount of time experienced by the weeble (or their watches, when they try it) is always an odd number of minutes because the object has to make an odd number of trips to end up at point B. In the movie they get 1347 minutes on the one time they try it.
- If the weeble is put in at point B, while the box is powering down, and then removed again at point B, the weeble must make an even number of trips. In the movie they get 1334 minutes on the one time that they try it.
Almost immediately once they discover this, Aaron and Abe reason that if they could create a device which had some degree of intelligence, like a programmable miniature robot or something, they could program the object to measure the amount of time that has passed and then spontaneously trigger the collapse of its own waveform from inside the box, thus selectively exiting the box at point B or point A. From this they imagine a box where the robot is inserted at 12:01pm, point B, waits for one minute, climbs out of the box at point A (12:00pm), and thereby travels backwards in time.
Aaron and Abe skip this step entirely and jump straight to building coffin-size boxes so that they can travel through time themselves.
Operating the box
When the box is turned on the field builds gradually. During this build-up there is a period during which the box is still turned on but the field inside it is still weak enough that an object can exit or enter it, and thereby drop into the indeterminate channel. This narrow window of a few minutes is point A. The channel persists until the box is switched off later. When this is done, the field diminishes and again there is a window of a few minutes in which an object can enter or exit the box safely. This is point B.
Operating the box becomes a matter of timing and preparation. The operator, let's say Aaron, turns on the box at the time he wishes to exit and then walks away so that he is not present for window A. Later, Aaron returns to the box and switches it off. As the machine powers down, window B opens, and he climbs inside. Time passes and the machine switches off entirely. At this point, instead of exiting, Aaron "bounces" off the far end of the channel and begins to loop around backwards in time instead of forwards. Because time is now running backwards, the box appears to power up again from Aaron's point of view. Aaron waits until the box appears to power down again (in reality, this is point A, where it powered up), and then exits the box through window A in a timely fashion. He is now Aaron-2.
Aaron-2 must avoid Aaron-1 for the rest of the day. Why? Because if they interact with one another, or the course of events of the day is altered in any way, then it would be possible that Aaron-1 would never get into the box at window B. This would not rupture the timeline or anything stupid like that, but it would mean that there are two Aarons in this timeline. Permanently. And only one identity between them.
While inside the box, the operator has no way of knowing how strongly the field is actually running. Exiting the box during one of the safe windows, then, requires advance calculation of the precise amount of time that the box was running, and you have to take a stopwatch into the box with you to make sure you stay in for the right amount of time. In theory, you could do what the weeble does and stay in the box for months or years, making multiple round trips before exiting. Indeed, if you died inside the box, then eventually your dessicated corpse would emerge at point B, having spent subjective years going around and around the loop before exiting. However, this is not explored in the movie.
Actually, windows A and B are arbitrary. It is possible to enter or exit the box at any time. However, doing this is dangerous. The more strongly the field is running, the more dangerous it is to do this. This is why Aaron and Abe only do it when the field is weak, i.e. when the box has just started powering up or is just about to finish powering off, and why Aaron feels seriously ill after his first trip-- he gets jumpy and exits very slightly too early (or too late, from the viewer's point of view). In addition, entering or exiting the box is never truly safe because the field is always active to some extent. There is always a small static shock, and Aaron and Abe experience cumulative ill effects even though they always use the boxes carefully.
When the weeble is being experimented on, it is possible to put the weeble in the box, switch the machine on from cold (point A), switch it all the way off again (point B), and then remove the weeble afterwards. This way, the weeble enters and exits the channel when the field is at precisely zero and so there is no danger to travelling through in time in this way. However, this is impossible when travelling backwards in time. You have to enter the channel before it has completely switched off and exit it after it has begun switching on. Thus, travelling back in time is inherently more dangerous than travelling forwards in time.
The box has to be flooded with argon, an inert, harmless noble gas, to operate, which is why Aaron and Abe have to use oxygen tanks on their trips. Because the boxes are small, dark and quiet, they also function like conventional isolation tanks, which is why Aaron and Abe dream inside and feel weird afterwards. This is probably also due to the sedatives they take before entering. Either way, it has nothing to do with the fact that they are travelling backwards through time.
Structure of the Primer timeline
The topology of the timeline is a straightforward forking model. Each trip through time creates a new timeline, divergent from all previous timelines, in which different things can happen. However, Abe's precautions of non-interference are still a good idea, as this ensures that a time traveller's past self enters the box and departs the timeline on schedule. If they do not, there are two of the same person in the same timeline, which raises substantial practical problems.
Exactly what would occur if you shut off a working coffin while you knew somebody was inside it (i.e. because you watched them climb out at point A), is not remotely clear.