When a city widens a highway to fix congestion, traffic gets worse within a few years — not despite the widening, but because of it. Planners used to treat this as a paradox. It isn't one. The congestion was never purely a capacity problem; it was also a rationing mechanism, the thing quietly deciding how many trips a metro area's road network could absorb before people gave up, carpooled, moved closer to work, or just didn't go. Widen the road and you haven't removed that pressure. You've removed the thing that was managing it, and all the trips it was suppressing show up at once, dressed as new demand.
The same shape appears wherever a scarce or clumsy constraint has been quietly performing two jobs: the job everyone complains about, and a second, unadvertised job of holding something back. Rent control freezes prices and, as an unintended second function, discourages landlords from building or maintaining units they'd otherwise convert or neglect for profit — remove the freeze and both effects show up together, the housing supply loosening and the displacement accelerating in the same motion. A strict content-review process at a publisher is slow and everyone wants it gone, but it's also filtering out the flood of marginal, hasty, or bad-faith submissions that a faster process would have to face head-on. Prohibition-era liquor bans didn't eliminate the demand for alcohol; they suppressed the visible market for it while an invisible one absorbed the same demand at higher cost and greater violence — repeal didn't create new drinking, it relocated drinking from black-market channels back into daylight, along with all the behavior the black market had been quietly containing along with the product.
None of these are arguments for keeping the friction. Congested roads are bad. Rent freezes distort housing stock badly enough that most economists oppose them. Slow review processes waste real time. The point is narrower and less comfortable: when a system built its behavior around a constraint for long enough, the constraint stops being separable from the outcomes people actually wanted. It becomes load-bearing for something nobody assigned it to bear. Removing it doesn't return the system to some clean, unconstrained baseline — it exposes whatever the constraint had been quietly doing the whole time it was also doing the thing everyone hated it for.
This is why "just remove the bottleneck" so reliably produces a second problem that gets described as new, surprising, unintended. It's rarely new. It was there the entire time, sitting underneath the bottleneck, invisible only because nothing had ever gotten past the bottleneck to reveal it. The honest question before removing any long-standing constraint isn't just "what is this costing us" — it's "what has this been quietly holding back, and are we ready to meet it in daylight."