Network effects are usually described as a single force pulling in one direction: each new user makes the product more valuable to everyone else. Real platforms run two forces at once. Every new participant adds matching value — more people to trade with, learn from, or hire — and simultaneously adds congestion: more noise to filter, more competition for the same scarce attention, more load on whatever ranking or moderation system decides who gets seen. Growth is the net of the two, and the net differs by user.

A regional marketplace for freelance editors shows the split cleanly. It opens registration beyond its invite list, and the user count climbs steadily for two quarters. For a client posting one job a year, the change is pure gain: more bids, lower prices, faster turnaround. For the twenty editors who earned most of the platform's revenue, it is pure cost — the same volume of work now arrives buried under a hundred cheaper bids, their response rate falls, and the reputation score they spent three years building stops distinguishing them from newcomers. They do not complain. They quietly move their repeat clients off-platform, where the congestion does not exist.

The reason this goes unnoticed for so long is arithmetic. Congestion costs land hardest on the smallest, highest-value cohort, and every headline metric — total users, total listings, aggregate transaction volume — is a sum or an average across a base growing fast enough to swamp them. The dashboard shows a healthy curve for the entire period in which the supply side is deciding to leave. By the time retention drops far enough to move an average, the cohort that generated the liquidity is already gone, and the cheap users who replaced them cannot reproduce it.