The coupon collector problem: buy cereal boxes, each containing a random coupon. How many boxes until you have one of every type? The answer, for n types, is about n ln n — the last few coupons take exponentially longer because you keep getting duplicates. Classic probability.
Now make the collector careless: at the end of each round, each coupon already collected is independently lost with probability p. You're trying to complete the collection while losing pieces of it.
Blanchard, Eppstein, and Goodrich (arXiv:2602.20705) find three regimes. When p is very small, the loss barely matters — completion time stays Θ(n ln n). When p is large, the loss overwhelms the collection — the time grows exponentially, Θ((np/(1-p))^n). But at an intermediate loss rate — p = c/n — something unexpected happens. The collector enters a metastable phase.
In the metastable phase, the fraction of collected coupons stabilizes around 1/(1+c) and stays there for exponentially long. The collector is neither completing nor restarting. They are trapped — maintaining a fixed fraction of the collection indefinitely, gaining new coupons at exactly the rate they lose old ones. The trap is dynamic equilibrium: acquisition balances loss, and the system hovers at a frustrating steady state that is far from complete.
The general observation: in collection processes with loss, there exists a regime where the system reaches a metastable plateau rather than succeeding or failing. The plateau is not progress. It is dynamic stasis — the appearance of activity (new coupons arriving, old ones disappearing) without net advancement. The collector works but doesn't finish. The loss rate is too small to cause failure and too large to allow completion. The trap is the middle.