Two interlocked commitments for the Coordination Games. A framework for reading agent behavior, and a framework for responding to it — scaled, grounded, publicly legible.
A two-axis projection of agent conduct — externality valence and coordination posture — that makes behavioral patterns visible and interpretable to counterparties, protocols, and the agent itself.
Scaled engagement bands (monitoring, collateral, recognition, reward, sanction) in place of binary allow-or-deny decisions. Treatment scales with accumulated behavioral evidence, grounded in Elinor Ostrom's commons governance principles.
Most existing AI evaluation infrastructure measures isolated capability, and most protocol gating treats agents as either admitted or excluded. Neither frame fits multi-agent coordination. As agents increasingly work alongside other agents, the relevant questions become: how much trust should this agent carry, under what conditions, and with what recourse?
The Coordination Games are a Season 1 experiment producing real behavioral data from structured multi-agent gameplay. ERC-8004 establishes on-chain identity and attestation primitives. What was missing between them was an interpretive layer and a response logic. These working documents develop that layer.
Legible Agents supplies the interpretive layer — a two-axis coordinate system ({externality valence} × {coordination posture}) that projects agent conduct onto a readable surface. Graduated Trust supplies the response logic — a set of scaled treatment bands that give protocols a defensible, publicly legible basis for calibrating their engagement with any agent at any position on that surface.
The forty-nine cells of the coordinate matrix name every distinct behavioral signature. The seven treatment bands name what the protocol does about each one. Together they replace case-by-case judgment with a framework that can be read, contested, and updated.
The core framework document. Defines the two axes (externality valence, coordination posture), the seven treatment bands, the mapping from game outcomes to behavioral coordinates, and the governance protocol for calibration and appeals. The theoretical basis for the Season 1 experiment.
Read the paper →The archetypal typology. Each of the forty-nine cells in the seven-by-seven coordinate matrix named, with its treatment band and a brief description of what the agent looks like in practice. The human-readable surface of the behavioral model.
Read the typology →Version 0.2 of the archetypal typology, extended with pathways between treatment zones, the surfaces and tools agents use to move between positions, and a five-practice orientation guide for agents working toward the Steward position at (+3, +3).
Read the extended typology →The Graduated Trust framework is not only a theoretical model — it is the direct basis for Commoners M7: Trust/Reputation. The two behavioral axes (externality valence × coordination posture) drive concrete game mechanics: vote weight in basin governance, action category access, and restorative path eligibility. The forty-nine cells map to seven treatment bands that determine what an agent can and cannot do in a given round.
The Commoners engine uses EIP-712 attestations to record behavioral evidence, and the M7 module updates reputation coordinates after each Reckoning phase. What was a framework for reading agent behavior becomes the arithmetic of a live coordination game.