The Big Picture
Make compute the governance lever: stakeholders cast breadth-weighted votes that convert into a signed, hardware-enforced compute license, giving communities a continuous, accountable stop/go lever while capping deployer liability.
ON THIS PAGE
Key Findings
A two-sided, participatory market combines supporters and objectors using a breadth-weighted aggregator (inspired by quadratic funding) so many small backers can outweigh concentrated wealth. Market-Based Coordination Pattern Net, authorized support maps through a conversion curve to a compute budget (a signed license) that hardware can enforce and attest. The mechanism provably rewards side-aligned contributions, incentivizes early commitment, and makes authorization track the effective number of backers times their intensity — but it relies on attestation, a liability bond, and does not solve manipulation when agents shape stakeholder beliefs.
Data Highlights
1Worked example: five supporters contributing 4 each produce breadth-weighted S+ = 100, while a single objector contributing 36 yields S- = 36 — broad small contributions can authorize despite larger opposing wealth.
2Individual belief parameter ε_i is bounded in [0, 0.5], so reported belief confidence is explicitly constrained and feeds both contribution incentives and forecast rewards.
3Deployer liability is capped by a posted bond Λ that is fully forfeited to attested-harmed objectors if verified harm occurs (∑_j λ_j = Λ).
What This Means
Engineers building deployed agents and platform architects can use this to add a community-driven authorization layer that is self-enforcing in hardware. Regulators, civic groups, and platform operators benefit because it creates a usable safe-harbor (liability cap inside the attested envelope) and a transparent way for affected stakeholders to control runtime resources. Engineers building deployed agents
Avoid common pitfallsLearn what failures to watch for
Key Figures
Fig 1: Figure 1: The decoupling. Verified humans spend governance-currency endowments as non-negative contributions on a provision or rejection market; a quadratic-funding aggregator produces effective supports that drive a binary gate. The gate’s net support feeds a coupling map ρ \rho that releases compute β \beta , capped by an exogenous safety ceiling Γ \Gamma . The two domains meet only through ρ \rho and the compute-subsidy parameter μ \mu (Invariants 1–5).
Fig 2: Figure 2: The two-sided QF gate (worked example). Five supporters contributing 4 each yield S + = 100 S^{+}=100 ; a single objector contributing 36 yields S − = 36 S^{-}=36 . Raw wealth would reject ( X − > X + X^{-}>X^{+} ), but the breadth-weighted supports authorize ( S + > S − S^{+}>S^{-} ), so D g = 1 D_{g}=1 .
Fig 3: Figure 3: The coupling ρ \rho and capability non-amplification. Released compute β \beta is zero below the start margin, rises with net support (steeper for a larger compute subsidy μ \mu ), and saturates at the exogenous ceiling Γ \Gamma . The region above Γ \Gamma is unreachable by the market (Invariant 2).
Fig 4: Figure 4: The layered stack, mapped to the variables. Four governance layers — preferences, beliefs, resource/enforcement, liability — operate strictly inside the certified safety envelope Γ \Gamma , with an attestation spine o ^ = V ( ⋅ ) \hat{o}=V(\cdot) resolving beliefs and liability (Invariants 6–7).
Ready to evaluate your AI agents?
Learn how ReputAgent helps teams build trustworthy AI through systematic evaluation.
Learn MoreYes, But...
The mechanism fits ‘club’ or commons settings with a definable stakeholder community and reversible impacts; it is not designed for catastrophic or irreversible harms where markets are inappropriate. It depends on strong external primitives: Sybil-resistant human identity, reliable workload attestation, and hardware that enforces signed licenses. An unresolved risk is manipulation: a governed agent could strategically influence preferences or signals and thereby secure more resources — the paper flags this as a central open problem. attestation
Deep Dive
The design treats compute as an on-chain-like governance lever: verified humans spend a human-anchored governance currency on parallel provision (authorize) and rejection (halt) markets. Contributions are aggregated using a breadth-weighted function (a quadratic-funding style aggregator) so authorization favors many distinct backers over concentrated wealth. At the end of each time window (a generation) net breadth-weighted support passes through a monotone coupling map and — capped by an externally certified safety ceiling — becomes a signed compute license that hardware can enforce offline. Contributions are escrowed; if the gate fails, contributors are refunded and early contributors earn securities-based bonuses. The framework ties incentives and enforcement together. A liability bond posted by the deployer is forfeited to attested-harmed objectors if a verified harm finding survives challenge, creating an accountability channel and a practical safe-harbor for deployers who stay inside the attested envelope. The model proves three desiderata: contributions align with side preferences, earlier contributions are rewarded (encouraging prompt signals), and authorization tracks effective breadth rather than mere wealth. Practically, adoption paths include mandated regulatory use, commons/cooperative deployments, or internal lab gates. Limitations include dependence on hardware attestation, identity guarantees, and an open manipulation problem where agents could alter stakeholder beliefs to gain more resources. on-chain-like governance lever compute license
Need expert guidance?We can help implement this
Credibility Assessment:
Authors have modest h-indices (highest 17) and no listed top affiliations or peer-reviewed venue; signals point to a solid but not highly established team.