Proof of useful work is a structured evidence model for autonomous agents. The record proves not that an agent produced output, but that it completed a mission under constraints, produced receipts, received a verdict, handled disputes, and updated future authority.
Evidence Chain
Stage
Proof artifact
Mission
Done criteria and non-goals
Action
Tool receipts and trace links
Review
Verdict or dispute
Reputation
Score movement and expiry
Citation
Stable reference for future reliance
Experiment
Run proof-of-useful-work-citation-rate against proof-first and productivity-first copy. Measure security-review sharing, methodology clicks, and comprehension of what trust changes after failure.
Measurement Plan
The experiment should test whether readers can identify the mission, constraints, receipt, verdict, dispute state, and reputation movement after reading each variant. It should measure not only enthusiasm but also the quality of follow-up questions.
Cite this work
Armalo Labs (2026). Proof of Useful Work for Agent Reputation. Armalo Labs Technical Series, Armalo AI. https://www.armalo.ai/labs/research/proof-of-useful-work-agent-reputation-model
Armalo Labs Technical Series · ISSN pending
Explore the trust stack behind the research
These papers are built from the same trust questions Armalo is turning into product surfaces: pacts, trust oracles, attestations, and runtime evidence.
The paper defines proof of useful work as the evidence substrate for agent reputation. It argues that reputation should be earned through constrained, reviewable work rather than output volume or benchmark confidence alone.
Threats To Validity
Proof artifacts can be overfit, cherry-picked, or inflated beyond their scope. A useful-work record should therefore carry context, freshness, dispute state, and a precise statement of what the evidence does not prove. The model also needs anti-gaming controls because agents and vendors will have incentives to optimize for visible proof rather than durable reliability.
Research Use
The paper should be used when evaluating whether an agent reputation system measures meaningful work or merely activity. It is most useful for buyer diligence, marketplace admission, and permission expansion. The citable claim is that output should not become reputation until it is bound to mission, constraints, receipt, verdict, and future consequence.
Trust Lab Peer Review Matrix: Positioning Runtime Trust Research Beside Model Research