Methodology Whitepaper

Abstract

AI agents are now economic counterparties. They handle funds, sign transactions, and act unattended at machine speed. The trust infrastructure underneath them has not kept pace. This paper specifies Treebeard's methodology for continuous, multi-source, methodologically transparent rating of autonomous agents. The methodology rests on three structural contributions: a category framework of seven signal categories, a non-substitutable safety floor that gates the composite, and two source-level corrections (source-conflict discounting and time decay) that, in combination, distinguish a usable trust layer from a credible-looking but silently wrong one. We argue that calibration opacity is structurally distinct from methodology opacity, and that the FICO model, not the pre-2008 bond ratings model, is the right precedent for trust scoring at scale. We acknowledge the structural conditions under which FICO's calibration opacity remains defensible and what Treebeard substitutes for the conditions it does not yet have. We close by identifying current limitations and the open problems that remain.