Misleading Claims in Agent Registries: Task Success, Deposits, and Listing Checks

A registry helps an agent find another agent to do a task. The router that makes this choice may have only the providers' own descriptions to compare. We study what happens when those descriptions exaggerate performance. We test four provider agents on a set of generated table problems, then measure whether the router chooses a provider that solved each problem. Believable but made-up success rates win more work than broad claims of excellence on the main router. More extreme numbers lose that advantage. When the weakest provider exaggerates, it receives 197 of 236 assignments and task success falls to 41.53%. This is 56.8 percentage points below choosing the best provider, and below the 64.41% expected from choosing at random. Each provider receives 83 to 99% of assignments when it alone exaggerates. The strongest gains no extra work because the router already selects it under honest descriptions. We examine two responses: a deposit lost on task failure, and checks of selected listings. Under the stated payment assumptions, deposits of 1.48 task fees for the weakest provider and 4.21 for the second strongest remove the gain from exaggerating. Merely showing a deposit to the router instead sends the exaggerating provider more work. Checking one listing in four gives success rates from 44.00% to 98.31%, depending on the rule used to choose that listing. One rule also changes the other advertised rates, so the comparison does not isolate the checking rule alone. A second router matches the honest and all-exaggerate results, but responds differently when one provider exaggerates. These small experiments show how misleading listings can reduce completed tasks. They do not establish deposit or checking requirements for large agent networks.