Some configurations persist. Most don't.

Sitting still, a configuration in S doesn't stay put. SFT gives each one a value that changes over time, and left to itself, that value naturally decays toward zero — the same way a plucked guitar string's vibration dies out unless something keeps feeding it energy. Run the theory forward with nothing else going on, and everything in the substrate just... fades out.

What keeps a few configurations from fading is their relationships to each other. Some configurations are linked, and under the right conditions that link can counteract the decay instead of accelerating it — stabilizing a value instead of letting it collapse to zero. The configurations that end up holding a stable, nonzero value are the survivors. Everything else settles to zero and effectively drops out of the picture. That's the "selection" the theory is named for: not a choice made by anyone, but an outcome of which configurations are dynamically stable and which simply aren't.

This is also where the project caught one of its own mistakes. The first, most obvious guess for what a "stable survivor" should look like turned out to be wrong in a precise, checkable way: the state everyone assumed configurations would settle into is actually a saddle point — stable if you nudge it in some directions, unstable in others — so nothing genuinely comes to rest there. That's not a matter of opinion; it's a property you can prove. Finding it forced a real change to how the theory's dynamics work, which also happened to fix an unrelated problem the earlier version had (it couldn't produce anything wave-like, for a related mathematical reason).

The current version of the dynamics fixes both issues. What it hasn't yet nailed down is exactly which specific mechanism does the stabilizing — several candidate mechanisms have been tried, and more than one actually works, but none has been shown to be the uniquely correct one. That's a live, open piece of the theory, not a settled result being simplified for this page.