The general idea is to iterate while changes are still propagating through the circuit; here's some pseudocode:
changedNets.add(initial stimulus)
for each net in changedNets
changedNets.remove(net)
for each node in nodes(net)
for each output in outputs(node)
evaluate output
if output changed
changedNets.add(output)
if state cycle detected
break
The above algorithm reasonably accurately models what happens in a real digital circuit: changes propagate through each node until the whole circuit reaches a steady state. You can even add visualisations to show each step of the propagation as it takes place, and more elaborate versions can take into account different propagation delays (i.e. the output of each node can change at a different time relative to its input, so you process them in a queue ordered by that.)To use my example circuit, with both inputs low initially: when one of the inputs of one NOR is set high, in the next step the output goes low, then the other NOR's output goes high, and the input of the first NOR goes high. This doesn't change anything else, so the circuit reaches equilibrium. When you set that input back down, the NOR already has the other input high, so it doesn't change state and equilibrium is already reached.
Cycle detection can be as simple as a hardcoded iteration count, or something a bit more accurate and useful:
https://en.wikipedia.org/wiki/Cycle_detection
Interesting fact: A true cycle-detection algorithm will theoretically let you simulate a whole CPU running any Turing-complete program, if you clock it with an "inherently unstable" ring oscillator...
The point of my exercise was to test a basic functionality. If you cannot simulate that, the feedback loops which arise naturally in more complex circuit designs won't work either. It's like the "fizzbuzz" of logic simulation.
Edit: maybe it should be called a "logic circuit" instead of a "digital circuit".
https://i.imgur.com/V6yXpMl.png
The gates themselves in isolation work fine, but connecting them together doesn't. I even tried "bootstrapping" the circuit by breaking the storage loop and setting the two gates' inputs so they're all consistent first before connecting them together, but then when I disconnect the bootstraps it immediately returns to the above arrangement.
IMHO having trouble with something like that means the simulation engine is not quite usable yet...
Edit: the above circuit works fine in Logisim. Also, as another minor tangent, I think the flip-flop is so much easier to understand when it's drawn to emphasise the storage loop, i.e. the two NORs essentially become inverters when one of their inputs is 0, and they act in positive feedback to hold the bit value; why almost every other diagram insists on crossing the wires is perplexing, since it only serves to confuse and complicate.