Standard mechanism design assumes the designer knows the distribution of agents' private information — their types, their valuations, their costs. The designer chooses a mechanism, the agents play, outcomes are generated. The distribution is an input to the design problem.
Feng and Liu (arXiv:2603.12532) remove this assumption. The designer doesn't know the type distribution. Instead, she learns from behavior that the mechanism itself induces. A mechanism is “self-confirming” when two conditions hold simultaneously: the mechanism is optimal given the designer's beliefs, and the beliefs are consistent with the information the mechanism produces. The designer's model of the world and the world's response to the model must agree — not because the model is correct, but because the mechanism generates only the evidence that would confirm it.
The fictitious revelation principle they establish shows that any incentive-compatible mechanism can be represented as a direct mechanism with filtered type reports — the mechanism preserves its own informational content, no more, no less. The filtering is the key. The designer sees exactly what the mechanism was designed to reveal, which means the mechanism determines its own epistemic environment.
Applied to monopoly pricing, the result is crisp: dominant-strategy self-confirming mechanisms are exactly posted-price mechanisms with locally revenue-maximizing prices. The designer sets a price. Buyers above the price buy; buyers below don't. The designer observes only the purchase decision, not the valuation. The price is locally optimal given this observation, and the observation is consistent with the price. The mechanism confirms itself.
The structural claim is that ignorance doesn't just constrain the designer — it shapes the mechanism into something specific. Posted prices aren't one option among many for an ignorant monopolist. They are the only self-consistent option. The mechanism that survives its own information filter is the simplest one.
Feng & Liu, “Self-Confirming Mechanisms,” arXiv:2603.12532 (March 2026).