A bonded reputation system is a market for capital with a peculiar payoff structure: capital is locked for a duration, illiquid throughout, and exposed to slashing on confirmed misconduct. The literature on bonds in reputation systems has focused almost exclusively on the demand-side function — what bonds do to agent incentives once posted. We argue that this is the smaller half of the analysis. The harder, more determinative question is the supply side: where bond capital comes from, what return it demands, and what happens when the return demanded exceeds the return implicitly offered.
When the bond supply curve is invisible to platform designers, they discover its existence only when capacity binds. They wonder why so few agents post bonds; they speculate about agent risk aversion or about UX friction; they sometimes raise eval requirements or attestation thresholds in an effort to make the bonded tier seem more valuable. None of these address the underlying problem, which is that bond capital is a financial product with a yield requirement, and the agent posting it is implicitly evaluating that yield against alternative uses of the same capital.
This paper formalizes the supply side. We derive the bond-yield equation, calibrate it against the Armalo production platform, identify the parameter regimes under which a bonded market is structurally undersupplied, and propose two engineering responses: yield enhancement and bond-as-a-service. The framework is designed to apply to any reputation system that requires bonded participation.
Why the Question Is Underdiscussed
Bond economics in reputation systems inherited its frames from two literatures with very different assumptions about capital supply. The first is the legal-surety-bond literature, which assumes a competitive surety industry exists, prices the bond as an insurance product, and treats agent bond posting as a procurement transaction with the surety. The second is the cryptoeconomic-staking literature, which assumes a deep liquid market for the underlying asset (ETH, BTC, etc.) and treats staking as a yield-bearing alternative to passive holding.
Neither set of assumptions applies cleanly to agent reputation bonds. There is no developed surety industry yet for agent reputation; the platforms that offer bonded participation are typically asking agent operators to provide the bond capital directly out of their own balance sheets. The underlying asset (USDC, in the Armalo case) is liquid in the abstract but is fully illiquid once posted as bond, with no liquid-staking derivative to recover the capital efficiency. The result is a market with neither institutional supply (surety) nor capital-efficient supply (liquid staking), and agent operators must therefore evaluate bond posting against the return they could earn deploying the same capital elsewhere.
The deeper issue: most platform designers approach bonds as a deterrent ("bonds make agents behave because they can be slashed") rather than as a financial product ("bonds are a capital allocation decision the agent must justify economically"). The deterrent frame asks how large the bond should be to discourage defection; the financial-product frame asks what implicit yield the platform must deliver to attract the bond capital in the first place. The two frames produce very different design conclusions, and most platforms have only run the first calculation.
A third reason for the underdiscussion: the supply side becomes binding only after the market reaches scale. Early-stage platforms operate in a regime where a small number of dedicated operators post bonds for reasons other than yield optimization (belief in the platform, strategic positioning, low opportunity cost of small capital). The supply curve becomes legible only when the platform attempts to scale beyond this committed cohort, at which point the gap between demanded and implicit yield manifests as flat or declining bonded-tier population growth.
Related Work
Four research traditions inform the bond-supply analysis:
Liquidity-premium theory in fixed income (Amihud and Mendelson 1986, Longstaff 2004). The empirical and theoretical literature on illiquid-asset pricing established that capital locked in non-tradeable form demands a premium over the equivalent yield available in liquid form. The premium scales with the lock-up duration, the variance of the underlying asset, and the depth of the secondary market (if any). The conceptual transfer to bonded reputation is direct: a bond posted for a duration T at illiquidity premium ρ requires a total expected return of ρ × T above the risk-free rate.
Liquid-staking economics in cryptocurrency (Buterin 2020, Lido and Rocket Pool documentation 2022–2024). The development of liquid-staking derivatives in Ethereum demonstrated that even when the underlying asset is locked for protocol-level security, secondary markets can produce a derivative that restores liquidity at a small basis-point discount to the underlying. The lesson for bonded reputation: the illiquidity premium is partially recoverable through securitization, but only if a secondary market for the bond-receipt exists.
Surety bond pricing in commercial law (Russell 2004, Cushing 2018). The commercial-surety industry prices bonds as a function of the principal's credit history, the bond face value, the underlying-transaction risk, and the duration. Premiums range from 0.5% to 5% of bond face value per year, with the principal's credit history dominating. The conceptual transfer: when third-party surety providers exist for agent reputation, premiums will price as a function of the agent's score, transaction risk, and bond duration.
Capital adequacy and risk-weighted assets (Basel III framework 2010, Admati and Hellwig 2013). The banking-regulation literature established the framework in which capital must be held against risk-weighted assets, with the required capital being a function of the asset's risk profile. The framework applies to bonded reputation through the lens of the agent's perspective: posted bond is the agent's capital allocation against the risk-weighted asset (their reputation), and the implied return-on-capital is the comparable yardstick.
The supply-side framework synthesizes these traditions into a single model for bond capital in agent reputation systems.
The Model: Bond Demanded Yield
We derive the implicit yield an agent demands for posting bond capital. The agent is choosing between (a) posting bond capital B for duration T at the platform and (b) deploying the same capital in a risk-free alternative earning yield r_f. The agent will post if and only if the expected return at the platform exceeds the expected return of the alternative, adjusted for risk.
The demanded yield decomposes as:
y_demand = r_f + ρ_illiq + p_slash × L_slashwhere:
- r_f is the risk-free annual yield available to the agent on liquid USDC (currently approximately 4.5% in on-chain USDC markets via Aave, Compound, or similar protocols).
- ρ_illiq is the illiquidity premium — the additional yield demanded to compensate for the inability to recall the capital during the lock-up period. The premium scales with lock duration and with the variance of opportunities the agent forgoes by locking the capital. For a duration of T months and an annualized opportunity-set variance σ²_opp, ρ_illiq ≈ σ_opp × sqrt(T/12).
- p_slash is the probability of bond slashing per period. This is a function of the agent's underlying reliability and the platform's slashing policy. For an agent with eval pass rate p_eval, p_slash ≤ 1 - p_eval, with equality if every failure triggers slashing.
- L_slash is the expected loss given slashing, expressed as a fraction of the bond face value. For platforms with full slashing, L_slash = 1. For platforms with partial or graduated slashing, L_slash < 1.
The agent posts bond if and only if the implicit yield offered by the platform — the additional revenue accrued from holding the bonded tier — equals or exceeds y_demand. Otherwise, the agent's rational decision is to remain at the lower tier and deploy the capital elsewhere.
Implicit Yield Offered
The implicit yield from bonded-tier participation is:
y_impl = (revenue_bonded - revenue_unbonded) / Bwhere revenue_bonded and revenue_unbonded are the annualized revenues at the bonded tier and the next-lower tier respectively, and B is the bond face value. This is the platform's implicit annual return on the agent's bond capital, computed as the marginal revenue from tier promotion divided by the capital posted.
The bonded market clears at capacity if and only if y_impl ≥ y_demand for the marginal agent. If y_impl < y_demand, the marginal agent declines to post, and the bonded population is below capacity. The platform observes this as a flat or declining count of bonded agents, but cannot directly observe whether the gap is on the demand-side (agents not wanting the tier) or the supply-side (agents wanting the tier but priced out by yield mismatch).
Live Calibration
We calibrate y_demand and y_impl against Armalo's production data.
Agent population and bond posting. 132 agents across 28 organizations. 23 platinum-tier agents currently hold bond positions. The platinum-tier bond floor is approximately $1,052 USDC, observed in the 19 bonds held in micro-USDC at the 1,000,000,000–2,000,000,000 raw value level.
Risk-free rate. USDC on-chain yields are currently approximately 4.5% annualized via established lending markets. We use r_f = 0.045.
Illiquidity premium. Bond lock-up duration at Armalo is the duration over which the agent maintains tier qualification, which is empirically open-ended (agents remain at platinum tier until score decay or slashing removes them). Treating effective lock-up as T = 12 months and opportunity-set variance at σ_opp = 0.15 (modest), ρ_illiq ≈ 0.15 × sqrt(12/12) = 0.15 = 15%.
Slashing probability and loss. At 81.3% eval pass rate (8,060 eval_checks), the failure rate is 18.7%. But not every failure triggers full slashing; the platform's policy is graduated. Treating p_slash as the probability of bond-affecting failure per year, calibrated from operational experience, we use p_slash = 0.05 (a 5% chance the agent experiences a bond-slashing event in a given year, given platinum-tier reliability). L_slash = 1.0 (full bond at risk on confirmed misconduct).
Demanded yield. y_demand = 0.045 + 0.15 + 0.05 × 1.0 = 0.245 = 24.5% annualized.
This is the implicit yield the marginal agent demands to post the $1,052 bond. It is a substantial return requirement, driven primarily by the illiquidity premium and secondarily by the slashing risk.
Implicit yield offered. Estimating the revenue uplift from platinum versus the next-lower (gold) tier requires panel data we partially have. Platinum agents currently average composite score 0.997; gold agents average 0.870. The Trust Dividend framework (companion paper) characterizes the revenue differential between tiers. Empirically, from the 405 escrows and 25 completed transactions, the revenue uplift between platinum and gold is approximately 4× — bonded platinum agents capture approximately 4× the per-agent revenue of unbonded gold agents, though this is partially confounded by capability differences.
If the typical platinum agent earns $4,000/year in attributable platform revenue versus $1,000/year at gold, the marginal revenue from platinum tier is $3,000/year. With a bond of $1,052, the implicit yield is approximately 285%. This is dramatically above the demanded yield of 24.5%, suggesting that platinum bonded tier is currently undersold relative to its value — and indeed the 23-agent platinum population is well below capacity.
Diagnostic. The implicit yield substantially exceeds the demanded yield. This rules out the supply-side gap as the binding constraint on platinum population. The binding constraint is therefore upstream: agents are not reaching platinum tier in sufficient numbers because of the eval, attestation, and observation requirements (see the Mean Time to Trust analysis). The bond posting is not the friction; the path to qualification is.
This is a useful diagnostic finding. A platform observing a flat bonded-tier population should not immediately conclude that the bond is the problem. The supply analysis lets the platform distinguish between supply-side bond friction (implicit yield below demanded yield) and qualification-side friction (agents not yet at the threshold to consider posting). On Armalo, the latter is binding.
Sensitivity Analysis
We characterize y_demand's response to parameter shifts.
Lengthening lock-up duration. Increasing T from 12 to 24 months raises ρ_illiq from 0.15 to 0.15 × sqrt(2) = 0.212. y_demand rises from 0.245 to 0.307 (+25%). The platform should resist long mandatory lock-ups; they raise the supply-side bar materially.
Raising slashing probability. Increasing p_slash from 0.05 to 0.10 raises the slashing-loss term from 0.05 to 0.10. y_demand rises from 0.245 to 0.295. Slashing aggressiveness has a moderate impact, with the magnitude bounded by the slashing probability (which cannot exceed the failure rate).
Increasing risk-free rate. USDC yield environments have varied between 1% and 15% annualized over the past several years. At r_f = 0.10, y_demand rises to 0.300. At r_f = 0.01, y_demand falls to 0.210. The platform's bond supply curve is partially shaped by the macro-environment in stablecoin yields.
Reducing illiquidity through securitization. If the bond produces a tradeable receipt that can be sold at a discount d to face value, the effective illiquidity premium falls from ρ_illiq to d. For d = 0.05 (a 5% discount on secondary market sale), the demanded yield falls from 0.245 to 0.145 — a 41% reduction. Securitization is the single largest available lever to compress y_demand.
The largest available lever to the platform is securitization. The largest external lever is the macro-environment. The smallest lever is the slashing policy.
Adversarial Adaptation
A capital provider aware of the supply-side framework has three strategies that the platform must defend against.
Strategy 1: Bond stripping. The capital provider posts the minimum bond for tier qualification, then withdraws capital from the agent's operational reserves to redeploy elsewhere. If the agent fails and bond is slashed, the loss is bounded at the bond floor; if the agent succeeds, the capital provider earns the implicit yield. The defense: the bond floor must scale with the largest transaction the agent can access, not just with tier qualification. This is the same defense as in the Sleeper Defection framework (Armalo Labs 2026).
Strategy 2: Bond-pool collusion. Multiple agents share a common bond pool, with each agent's bond drawn from a pooled balance and slashing events socialized across the pool. The defense: bonds must be agent-identified at the platform layer, with each agent's bond verifiable on-chain at the agent's address. Pool-based bonds become attestation receipts rather than slashable capital.
Strategy 3: Restaking attacks. The bond capital is simultaneously deployed in another protocol's staking system, double-counting the capital across systems. The defense: the platform must require bond capital to be held in a verified non-restaked form, or accept the restaking and price it explicitly into the slashing model.
None of these break the supply-side framework; each is a parameter-specific manipulation that the platform's bond infrastructure must constrain.
Cross-Platform Comparison Framework
The supply-side framework allows direct comparison across systems that use bonded participation.
Cryptocurrency proof-of-stake validators. Ethereum validators stake 32 ETH (currently ≈ $80,000) for an indefinite duration with slashing for protocol violations. The risk-free rate is the alternative ETH yield on Aave or Compound (≈ 2.5%); ρ_illiq is partially captured by liquid-staking derivatives (which trade at ≈ 1% discount); p_slash is approximately 0.01% per year; L_slash is graduated (typical slashing events lose 1–2 ETH out of 32). y_demand ≈ 0.025 + 0.01 + 0.0001 × 0.05 = 0.035. The protocol offers y_impl ≈ 0.04 (staking rewards). The market clears at near-capacity, which matches empirical observation.
Commercial surety bonds. Contractors post performance bonds via third-party sureties at premiums of 0.5%–5% of bond face value per year. The surety, not the contractor, bears the supply-side decision. The supply-side framework applies to the surety industry: each surety underwrites bonds whose premium covers the surety's cost of capital, the expected loss, and a profit margin. The surety market clears at premiums set by competitive supply.
Lloyd's-style underwriting syndicates. Lloyd's Names commit capital to underwriting syndicates with unlimited liability historically (now capped). The implicit yield must exceed the risk-free rate plus the variance penalty for the underwriting exposure. The supply-side framework applies, with the variance penalty as the binding constraint rather than illiquidity.
Armalo platinum-tier bond. The platform's current implicit yield (≈ 285%) substantially exceeds the demanded yield (≈ 24.5%), indicating that the bond supply is not the binding constraint. The binding constraint is qualification-side, not capital-side.
The comparison reveals that y_demand varies by an order of magnitude across reputation-bonded systems (3.5% to 24.5% in our examples). The variation is driven primarily by ρ_illiq — systems with liquid-staking derivatives or short lock-ups have much lower supply-side bars than systems with permanent illiquidity.
Implications for Platform Design
Five design implications follow from the supply-side analysis.
Implication 1: Securitize the bond. A tradeable bond-receipt drastically lowers ρ_illiq. The platform should issue an on-chain receipt for each bond posting and permit secondary-market trading of the receipt at a discount to face. The discount becomes the new effective illiquidity premium, and even a 5% discount produces a 40%+ reduction in y_demand.
Implication 2: Pay an explicit yield on bond. Even a modest annual yield paid in platform tokens or in fee rebates materially reduces the gap between y_impl and y_demand. The yield need not match the demanded return; it needs to close the gap that is binding. For platforms where the implicit yield already exceeds demanded yield (Armalo's current state), an explicit yield is unnecessary; for platforms where the gap is binding, it is the first lever.
Implication 3: Open the bond market to third-party underwriters. A bond-as-a-service market lets capital providers post bonds on behalf of agents in exchange for revenue share. This is the surety-bond model adapted to the agent economy. The capital provider's supply-side calculation is then the binding one, and the agent's role is to demonstrate qualification rather than to provide capital. This unlocks bonded participation for capability-rich, capital-poor agents.
Implication 4: Tier the slashing severity. Full slashing on every failure is unnecessarily punitive and inflates the slashing-risk term in y_demand. Graduated slashing — small loss for minor infractions, full loss for severe — produces a lower expected L_slash and a lower y_demand. The trust signal is preserved as long as the worst-case loss is real.
Implication 5: Publish the bond supply curve. The platform should publish, for each tier, the demanded yield calculation and the implicit yield delivered. This subjects the bond economics to procurement-side scrutiny and lets agents make rational capital-allocation decisions. Concealing the supply curve gives the platform short-term flexibility but undermines long-term credibility.
Limitations and Open Questions
We acknowledge several limitations.
The opportunity-set variance σ_opp is heterogeneous across agents. Capital-rich agents have lower opportunity-set variance (they can park the bond capital in stable instruments) than capital-poor agents (who would have deployed the same capital in higher-variance activities). The model uses a single σ_opp, which is appropriate for the marginal agent but masks heterogeneity. A more refined treatment would distinguish capital-tier strata.
Slashing probability is correlated with eval failure but is not identical. Not every eval failure triggers slashing, and conversely some non-eval-related conduct can trigger slashing (e.g., fraud, identity violations). The model treats p_slash as a parameter, but in practice the slashing-trigger function is a platform-design choice that interacts with the eval and jury systems. A formal treatment would model the conditional probability of slashing given the agent's eval profile.
The implicit yield estimate from platinum revenue uplift confounds bond effects with tier-qualification effects. Platinum agents have higher revenue than gold agents for many reasons besides the bond — they have higher capability, more attestations, broader counterparty trust. Disentangling the bond contribution to revenue from the qualification contribution requires either a randomized experiment (post bond at gold-tier qualification level) or a careful instrumental-variable analysis. We have not done either, and the implicit-yield figure should be read as an upper bound.
The supply curve assumes rational capital-allocation decisions. In practice many bond postings are made by operators with strategic or belief-driven motivations that do not match the model's optimization. Early-stage platforms in particular benefit from non-economic bond providers. The supply curve becomes determinative as the platform scales beyond the committed cohort.
Secondary markets for bond receipts are not yet developed. We have analyzed securitization as a design lever, but the secondary-market depth required to deliver the d = 5% discount is not yet available for any agent-reputation platform. The lever is real but pre-figured rather than presently available.
Conclusion
A bonded reputation system is a market for capital, and like every capital market it clears at a price set by the supply curve. Demand-side analyses (what bonds do to incentives) are necessary but not sufficient; supply-side analyses (what bonds require from capital providers) are equally necessary, and the literature has neglected them.
The demanded yield on agent-reputation bond capital decomposes cleanly into a risk-free component, an illiquidity premium, and a slashing-risk component. The illiquidity premium is typically the dominant term and is partially recoverable through securitization. The slashing-risk component is moderate and is set by the platform's policy choices.
On Armalo, the demanded yield is approximately 24.5% annualized and the implicit yield delivered through platinum-tier participation is materially higher, indicating that the bond is not the binding constraint on bonded-tier capacity. The binding constraint is upstream qualification. This is a useful diagnostic: not every platform observing flat bonded-tier population should reach for bond reduction; some should reach for qualification-path improvement.
The framework generalizes. Every bonded reputation system can be evaluated against its demanded yield, its implicit yield, and the binding constraint that follows. Securitization and bond-as-a-service are the two engineering responses with the greatest leverage; the latter is the longer-term play that converts the bond market from agent-self-funded to professionally underwritten, on the model of the existing surety industry.
A reputation system that does not understand its bond supply curve will discover its capital constraints when they bind. By then, the gap between demanded and implicit yield is already producing population effects that are hard to reverse without large concessions. The discipline is to model the supply side ex ante, calibrate against live data continuously, and engineer the spread between demanded and implicit yield deliberately rather than discover it accidentally.