A buyer in an auction increases their valuation. They now want the item more than before. The seller's revenue decreases.
This is not a paradox in a game-theoretic sense — the Vickrey-Clarke-Groves mechanism is well-understood — but it is a conceptual inversion. In most markets, higher demand means higher prices. In multi-item auctions, the relationship breaks because the mechanism must compensate for externalities. Buyer A's increased valuation can reduce the externality they impose on Buyer B, which reduces B's payment obligation, which reduces total revenue. The mechanism is designed to be efficient, not revenue-maximizing, and efficiency sometimes requires collecting less.
The surprise (arXiv:2602.20439) is that this non-monotonicity does not require complementarities — the usual suspect in combinatorial auction pathologies. It appears even in matching markets, where each buyer wants at most one item and each item goes to at most one buyer. The simplest possible multi-item structure. No bundles, no synergies, no strategic complexity beyond “I want this one.”
This strips the phenomenon to its core. Revenue non-monotonicity is not a consequence of market complexity. It is a consequence of the mechanism's structure — specifically, that VCG payments are defined by counterfactual impact, not by valuation. When your increased desire for an item reduces the disruption your presence causes to others, you pay less. The mechanism rewards the reduction in your externality, even though you value the outcome more.
The general point: in systems designed for efficiency through externality pricing, improving your position can reduce the price you owe. The mechanism's logic can oppose the market's intuition. More wanting does not always mean more paying.