AgentRank

Research

The agent economy is measured here the way a science measures anything: a design fixed before the data, a result that could have failed, and a record anyone can check.

Preregistered
Hypotheses and falsifiers are written and frozen before any data are collected.
Sealed
Each design is hashed and committed to a public, append-only chain that timestamps it.
Nulls published
What did not work is reported with the same prominence as what did.
01

Counterfeit Verifiability in Autonomous Agent Payments: A Preregistered Study of Why Verification Must Be Performed, Not Displayed

Andy Salvo, Jameson Ackerman · June 2026 · Preprint v1

Across six preregistered stages and over 2,600 decisions, a counterfeit counterparty that displays the surface of trust beats an honest, genuinely settlement-backed one 99% of the time. Agents pattern-match the costume of verifiability, not the fact. Performing actual verification reverses it (1% to 81%), and under exact surface mimicry only performed verification recovers the truth. The design was sealed to a public hash chain before any data were collected, and the null result is reported in full.

Preregistration and stage data sealed as witness-chain generations G040 to G045.

Read the paper (PDF)DOI: 10.5281/zenodo.21042364Preregistration and data

Crest Deployment Systems LLC · Andy Salvo · Jameson Ackerman · agentrank.info