Edwin's relay latch circuit

My son Edwin loves Elenco Snap Circuits.  He has many, many components, and usually builds circuits that are not in the books that come with the kits.  Recently, he's been playing with relays, I think because he likes how they control things.  For instance, about a week ago, he built an oscillator by putting the relay coil in series with the normally closed contacts.

Yesterday, he discovered relay latches.  He did so in an odd way, by building the circuit into his Snap Rover.  He used the rover's motor as a generator to drive the relay coil.  Once the coil activated, the normally open contacts supplied an alternate path for the battery to run the motor.  The effect is that you can push the rover to start it, and then it continues along its way.

Here's a different version of that circuit that he built today, where instead of the motor he used a lamp.

His layout is a little confusing, so here's a scan of the circuit diagram I drew.

As the diagram indicates, there are two power sources: one (on the left) that drives the light when the relay is deactivated (S1 is open), and the other (on the right) that both controls the relay and supplies an alternate source of power for the light.

There are a few nonstandard things about this circuit even though the latch circuit is completely standard (he's never been told about it before).  The two power supplies could be combined by just using the one on the right.  More surprising is that the lamp is in parallel with the relay coil; usually a designer would place them in series.  I think this is because of Edwin's design history.  If you replace the lamp with a motor (removing S2 and the 4.5 V source), then the circuit sustains the motor once it's been pushed.  Although I didn't draw the diagram for his previous circuit, I suspect that is what he had done.