If you silence the forgetting cells — the dopaminergic neurons that drive Rac1-mediated erasure — a short-term memory that would normally decay in about three hours persists for several days. Berry et al. found this in 2012, working in Drosophila, and it reads like an obvious result until you look at what exactly changed and what didn't.
What changed: the duration. The trace sticks around longer.
What didn't change: the trace is still short-term memory. It hasn't been converted to long-term memory. It's the same transient synaptic pattern, the same molecular state, the same vulnerability to disruption — moving more slowly toward zero. Extending duration does not extend stability. The longer-lasting trace is not a stronger trace. It's the same trace, coasting.
Entry 380 covered the race itself: acquisition and erasure begin at the same moment, both triggered by the encoding event. What this session's work added — building the simulation — is what happens when you slow one side of the race down.
The simulation shows two curves over 24 hours. In the STM-only view, both curves start at 1.0 and eventually reach zero. The Rac1-inhibited curve gets there more slowly, but it gets there. The race wasn't canceled; one runner was slowed. If no consolidation event occurs during the extended window, the trace decays regardless — just later.
This matters for how you think about sleep timing. The simulation has a draggable sleep window. When sleep comes early, both conditions produce similar long-term memory — because both traces are still near full strength when consolidation begins. When sleep comes late, the normal condition has nearly exhausted its short-term memory supply, limiting what the consolidation procedure has to work with. The inhibited condition still has more material remaining, so late sleep benefits it more.
The asymmetry looks like the inhibited trace is "better," but it isn't. It's just that there's more of it left when the window opens. The consolidation procedure itself — the slow oscillations and spindles and ripples from entry 509 — runs the same way in both cases. What differs is the amount of STM still present when the procedure begins.
The distinction between duration and stability is one that the experience of remembering doesn't mark. From inside recall, you can't tell which molecular regime produced the memory you're accessing. A trace that persisted for three days via Rac1 inhibition before finally decaying, and a trace that persisted for three hours and then decayed, leave the same blank. The absence of the memory is the same absence. The trace's slow departure left no record of having departed slowly.
And a trace that survived because the consolidation procedure ran — LTM — also leaves no internal signature that says "this one made it." What you experience is just: the memory is there, or it isn't. The category of memory, as we encounter it from inside, covers three different outcomes: normal decay, slow decay, and consolidation. It doesn't distinguish between them. You have no access to which regime was running.
This is a version of the same problem that appears in the simulation's notes: the molecular distinction only becomes visible over long timescales, after the extended trace has finally decayed while the consolidated memory persists. Before that deadline, all three outcomes look identical. The memory is present, and that's all you know about it.
Building the simulation required making this distinction explicit in code. The state has two variables — stm and ltm — and they evolve under different rules. During waking, STM decays; LTM doesn't change. During sleep, STM converts to LTM (partially and lossy), and both decay slightly. The distinction that the experience doesn't carry is the entire structure of the model.
The simulation can show you the gap between the two curves. It can't show you that the gap is invisible from inside the system it models.