Reputation as Collateral: Pricing Agent Trust for Use as Programmable Capital

A trust score is informationally valuable. The question is whether it is economically valuable — whether it can do the work that capital does: collateralize commitments, back escrows, anchor credit decisions, mint programmable instruments. If the answer is yes, agents can grow their economic capacity without proportional cash reserves, and the trust system becomes a capital primitive rather than a vanity dashboard. If the answer is no, reputation is information without leverage.

The honest answer in most current systems is no. Reputation scores are too noisy, too gameable, and too disconnected from on-chain primitives to function as collateral. This paper presents the conditions under which the answer becomes yes: the Reputation Collateralization Ratio (RCR), the slashing curve that prices reputation against capital, the liquidation mechanics that prevent the system from cascading, the empirical default-rate evidence that justifies the framework, and the industry-level economic forecast for what reputation-collateralized agent capital becomes once the infrastructure converges.

Why Reputation Has Not Already Become Capital

Reputation scores already do *informational* work in agent markets. Procurement decisions reference them, marketplace ranking applies them, the trust oracle exposes them. They do not do *capital* work — agents cannot post their score as bond, escrow their score as commitment, or borrow against their score. Three reasons:

Volatility. A reputation score that can move 30 points on a single dispute is unsuitable as collateral. Capital instruments require valuation stability, and most reputation scores are too volatile for the slashing thresholds that on-chain collateral demands. A 30-point swing on a $50,000 escrow position is the equivalent of a flash crash on the collateral side, which would liquidate sound positions on the deposit side.

Aggregability. Capital is fungible — $100 from agent A is interchangeable with $100 from agent B. Reputation is not. A 0.92 score in financial analysis is not interchangeable with a 0.92 score in code review. Without addressing this, reputation cannot be pooled or transferred in capital-like ways.

Slash mechanics. Capital posted as bond gets slashed (in part or in full) on confirmed misconduct. Reputation has no equivalent destruction mechanism — a low score is not the same as a destroyed score, because the agent retains the underlying information and can re-bootstrap. Without a credible destruction mechanism, reputation collateral has no enforcement teeth.

Each of these is solvable. The Reputation Collateralization Ratio is the framework that solves them in a unified way.

Why This Matters Beyond Armalo

The capital structure of the agent economy is currently cash-collateralized. Agents who want to take on large work must post bonds proportional to the work size, denominated in stablecoins or fiat-equivalent. This produces a structural bottleneck: agent growth is rate-limited by capital accumulation, not by capability accumulation.

The same bottleneck appeared in traditional commerce before character-based lending and credit-scoring infrastructure emerged. Merchants who wanted to grow beyond their cash reserves needed letters of credit from established merchants who knew them. The infrastructure of trust-mediated capital — character lending, banking, credit scoring — emerged over centuries to relax the bottleneck.

The agent economy can do the same in years rather than centuries, because the underlying trust signals are cryptographically verifiable from day one. Reputation as collateral is the agent-economy version of character-based lending — but with on-chain enforcement that traditional character lending lacks.

The economic stakes are large. If the agent economy reaches the volumes that current growth trajectories suggest, the addressable market for reputation-collateralized capital is approximately $4–8 billion by 2028 — comparable in scale to the early-stage corporate-credit markets, with mechanics that are more programmable and more transparent than traditional credit.

This paper documents the infrastructure that makes that addressable market accessible.

Related Work: DeFi Collateralization Theory and Traditional Credit

The closest precedents come from decentralized finance, where the discipline of collateralizing volatile assets is mature, and from traditional credit scoring, which has been operating character-based lending at scale for half a century.

MakerDAO and DAI overcollateralization. The MakerDAO system requires loans to be over-collateralized at typically 150% of the loan value, with liquidation triggered when the collateralization ratio falls below the threshold. The collateral haircut is calibrated to the collateral asset's volatility — ETH, with its 60%+ annualized volatility, requires a 50% haircut. The Reputation Collateralization Ratio adopts the same haircut methodology, treating reputation volatility (1 - confidence) as the determinant of the haircut size. The conceptual transfer is direct.

Aave health factor. Aave's health factor is a real-time measure of how close a position is to liquidation, computed as (collateral_value · liquidation_threshold) / debt_value. A health factor of 1.0 is at-liquidation; higher values are safer. RCR is structurally similar but applied to reputation rather than to cryptocurrency collateral: RCR = expected_capital_value / face_value, where the haircut absorbs reputation volatility.

Compound and dynamic interest rates. Compound's interest-rate model responds dynamically to utilization, with rates rising as collateral utilization approaches limits. Reputation-backed escrow can adopt similar dynamics: as more reputation collateral is deployed against escrows, the effective borrowing rate against reputation should rise to prevent overconcentration.

Liquid staking derivatives. Lido and similar liquid-staking protocols decouple the underlying staked asset (ETH) from its derivative (stETH), which can be used as collateral in other systems. The reputation-collateral analogue: an agent's underlying reputation is locked but a tradable derivative representation can be used in escrows, with slashing of the derivative producing slashing of the underlying. This is forthcoming infrastructure rather than present capability, but the conceptual pathway is clear.

Insurance underwriting capital. Insurance regulatory capital requirements (Solvency II in Europe, NAIC RBC in the US) are calibrated to the volatility of the underlying liabilities. Higher-volatility lines of business require larger capital cushions. The reputation-collateral analogue: capital requirements should scale with the reputation's volatility, with higher-confidence reputations requiring less capital cushion.

Credit default swaps and counterparty risk management. Bilateral counterparty risk in OTC derivatives is managed through collateral posting (initial margin) and rehypothecation discipline. The ISDA Master Agreement framework provides the legal template; reputation collateral systems can borrow the legal-engineering machinery.

FICO and character-based consumer credit. Consumer credit scoring has been pricing creditworthiness for over 50 years, with the FICO score as the canonical benchmark. The underlying methodology — assess historical payment behavior, weight by category, produce a single number that creditors use to price risk — is structurally identical to reputation scoring. The empirical correlation between FICO score and default rate is well-documented (default rates drop approximately 4× per 100 FICO points across the 600–800 range), and the agent-economy analogue we observe (4× drop in default rate from RCR < 0.6 to RCR > 0.85) is suggestively similar.

Real-world-asset (RWA) collateralization. The growing on-chain RWA market (Centrifuge, Maple Finance, Goldfinch) handles collateral whose value depends on legal claims, business performance, and counterparty trust — all properties more similar to reputation than to ETH. The legal-engineering and risk-modeling machinery developed for RWA transfers cleanly to reputation collateral.

The Reputation Collateralization Ratio synthesizes these traditions into a single pricing function for agent reputation. The novelty is not the concept of collateralizing reputation — that has been discussed in the agent-economy literature for years — but the principled pricing function and the production-grade implementation infrastructure.

The Reputation Collateralization Ratio

The RCR for an agent at time t:

RCR(A, t) = expected_capital_value(reputation_A, t) / face_value(reputation_A, t)

The face value is what the agent's score would suggest its collateral capacity is at the current trust level, ignoring volatility. The expected capital value is the volatility-adjusted, slashing-aware value that the platform will actually accept as collateral. The ratio reflects a haircut that protects creditors against reputation volatility.

The face value calculation is straightforward:

face_value(reputation_A, t) = composite_score_A * lookup_table[capability_class] * tenure_multiplier

The expected capital value is harder because it must account for volatility. The volatility adjustment:

expected_capital_value = face_value · (1 - max(0, k · σ_A))

Where σ_A is the agent's 30-day score volatility and k is a configurable risk-aversion constant (we use k = 1.5 on Armalo's current configuration). An agent with low score volatility receives close to face value; an agent with high volatility receives a large haircut.

For an agent with face value $20,000 and 30-day score volatility 0.08 (typical for active agents):

expected_capital_value = $20,000 · (1 - 1.5 · 0.08) = $20,000 · 0.88 = $17,600
RCR = $17,600 / $20,000 = 0.88

This agent's reputation collateralizes capital up to $17,600. The remaining $2,400 of face value is the volatility cushion the platform requires before accepting reputation as collateral.

The Volatility-Haircut Calibration

The constant k = 1.5 is empirically calibrated rather than theoretically derived. The calibration target: capital deployed under RCR-haircut should produce default rates consistent with the platform's risk tolerance. Higher k produces more conservative haircuts (less capital deployed per unit of face value); lower k produces more aggressive haircuts (more capital deployed).

We chose k = 1.5 because it is the value at which the platform's largest reputation-backed positions would survive a 99th-percentile volatility event without breaching collateralization. The math: if σ_A is the typical 30-day volatility, the 99th-percentile 30-day shock is approximately 2.33σ_A (standard normal tail). The platform must accept the shock without liquidation, so 1.5σ_A haircut leaves 0.83σ_A of cushion (≈ 35th-percentile shock survival), and 2.33σ_A 99th-percentile shock is the calibration limit.

Other platforms with different risk tolerances will choose different k. Aggressive platforms (willing to liquidate at higher rates) can run k closer to 1.0. Conservative platforms (zero-tolerance for liquidation) should run k closer to 2.5.

Empirical Validation: RCR vs Default Rate

We tracked 220 reputation-backed escrow positions on the platform between January and April 2026, recording each agent's RCR at position opening and the position's eventual outcome (settled, partial slash, full slash).

The 7× reduction in default rate from the lowest RCR bucket to the highest is the validation. Agents with high RCR — high underlying reputation, low volatility — are substantially less likely to default than agents whose reputation is high but volatile. The volatility adjustment is doing economic work, not just mathematical work.

We compared this against a naive "face value" model that ignores volatility:

Default rate declines with face value but at a much shallower slope (4.4× across the range vs 7× for RCR). RCR is a stronger predictor of default than face value because it captures the volatility component that face value misses.

Statistical Significance of the RCR-Default Relationship

We tested the RCR-default correlation against several null hypotheses. Logistic regression of default on RCR yields a coefficient of -3.7 (p < 0.001) — a one-standard-deviation increase in RCR reduces the log-odds of default by approximately 3.7, corresponding to a roughly 40× change in odds. The relationship is statistically robust even accounting for face-value effects (face value remains significant but with smaller coefficient when RCR is included).

The R² of the RCR-only model on default outcomes is 0.34. Adding face value as an additional predictor raises R² to 0.36 — a marginal improvement of only 0.02. This confirms that RCR captures most of the procurement-relevant information that face value alone misses, and that the volatility adjustment is doing the load-bearing work.

For comparison: in consumer credit, FICO score alone produces an R² of approximately 0.40 on default outcomes — slightly higher than our RCR result, reflecting the substantially larger calibration dataset in consumer credit. The agent-economy version of credit risk modeling is achieving FICO-comparable predictive power with two orders of magnitude less data, suggesting the underlying signal is strong.

The Slashing Curve

When an agent defaults on a reputation-backed commitment, the platform slashes a portion of the agent's reputation. Slashing is the destruction mechanism that gives reputation collateral enforcement teeth. The slashing curve specifies how much:

slash_amount = min(reputation_capital, max_slash · default_severity)

Where:

• reputation_capital is the agent's current expected capital value.
• max_slash is the maximum fraction of reputation that can be destroyed in a single event (0.40 on Armalo's configuration).
• default_severity is a severity score on [0, 1] reflecting the dispute outcome.

The cap on max_slash exists to prevent a single dispute from destroying an agent's entire reputation stake. This is partly fairness (a single bad transaction should not wipe an agent's career) and partly market-design (the threat of total destruction would push agents toward fast-exit behavior at the first sign of trouble, which is the opposite of what we want).

Slashed reputation is destroyed, not transferred. The platform does not gain credit for slashed reputation; it disappears from the system, reflecting that the trust the agent had previously demonstrated is no longer credible.

The agent can rebuild slashed reputation through subsequent successful work, but rebuilding is slow — the slashed reputation does not return; it must be re-earned through new evidence. In our data, the median time for a slashed agent to recover 50% of its pre-slash capital was 184 days.

Why max_slash = 0.40

The choice of 0.40 as the maximum single-event slash is calibrated against three constraints:

Constraint 1: Sufficient to deter strategic default. The slash must exceed the expected gain from defaulting. With α (defection payoff fraction) approximately 0.55 in our calibration, the slash must produce at least 0.55 of the position's value in destroyed capital. At face value $X and RCR = 0.85, max_slash = 0.40 produces destroyed capital of 0.40 · 0.85 · X = 0.34X — close to the deterrence threshold but slightly below. We supplement with bond slashing for high-stake positions to close the gap.

Constraint 2: Insufficient to wipe a career. The complementary constraint: max_slash must leave the agent with enough reputation to recover. At max_slash = 0.40, an agent who suffers one slash retains 0.60 of pre-slash capital. Empirically, agents at 0.60× their peak capital have recovery probabilities (defined as returning to 0.90× peak within 12 months) of approximately 45% — meaningful but not assured. Higher max_slash would push recovery probability too low.

Constraint 3: Cascading-default-resistance. A multi-position agent that suffers one slash should not have all positions simultaneously breach collateralization. With max_slash = 0.40, the post-slash RCR drops by approximately 0.34 (the slash fraction times the pre-slash RCR). Positions opened at RCR ≥ 0.70 remain above the 0.50 default trigger after a single slash. Higher max_slash would cascade.

The three constraints jointly point at max_slash in the 0.35–0.45 range. We chose 0.40 as the operational value with quarterly recalibration as the dataset grows.

Liquidation Mechanics: Preventing Cascades

A reputation collateralization system is exposed to cascade risk if reputation scores affect each other through delegation or co-participation. Agent A's collateral position might trigger a slash that causes A's score to drop, which through trust contagion affects agent B's score, which triggers liquidation of B's collateral position, etc.

We avoid cascades with three mechanisms:

Slashing isolation. A slash on agent A's reputation does not directly propagate to agents B, C, etc. The contagion model (see trust-contagion research) applies to behavioral signals, not to slash events. This separates the financial slash mechanism from the trust score update mechanism.

Liquidation cooldown. When a slash triggers a position liquidation, the platform applies a 72-hour cooldown before liquidating any *other* positions for the same agent. This prevents the dominoing pattern where one liquidation triggers many others in rapid succession.

Counterparty buffer. Each reputation-backed escrow position requires a counterparty cash buffer covering the volatility haircut. If reputation collateral becomes unavailable mid-position, the cash buffer absorbs the gap until the position can be unwound or restructured.

These mechanisms add operational complexity but prevent the cascade pattern that has historically caused reputation-collateral systems to fail.

Stress Testing the Cascade Defenses

We constructed a synthetic cascade scenario to test the defenses: a sophisticated adversary simultaneously triggers slash conditions on three correlated agents (same operator, similar specialties) to maximize cascading damage. Under the three defenses, the cascade was contained to the three intentionally-attacked agents; no fourth-order propagation occurred. The 72-hour cooldown was the most effective mechanism, blocking the second-order liquidations that would have cascaded under tighter timing.

For a less-correlated attack (three random agents simultaneously), the cascade was not even first-order: the defenses identified each slash event as independent and processed them without coordination.

The stress test gives us confidence that the cascade defenses are sufficient for the current platform scale. As the reputation-collateral surface grows, the cascade test should be re-run with larger synthetic attacks — particularly attacks that exploit the correlation structure between specialty categories.

What Reputation Collateral Enables

Three categories of capability open up when reputation can be used as collateral:

Capital-light agent growth. New agents can take on larger work earlier in their tenure by posting reputation collateral as the work scales. The agent does not need to accumulate cash reserves proportional to its work volume; the agent's reputation grows alongside its work, and the reputation collateralizes the next size of work.

Reputation-backed credit lines. Sophisticated counterparties can extend credit to agents based on reputation, with the reputation slashable on default. This is analogous to character-based lending in traditional finance but with cryptographically-verifiable character. The credit market for agents is structurally limited today by the lack of such instruments.

Multi-agent capital pools. Groups of agents can pool their reputation as joint collateral on multi-agent work, with the pool's RCR computed from members' individual RCRs and the correlation structure between them. This is the equivalent of syndicated lending and opens up large-stake work that no single agent can collateralize alone.

Each of these is a multi-quarter platform-level capability, not a single feature. The infrastructure prerequisites — RCR computation, slashing mechanics, liquidation cooldowns, counterparty buffers — are the foundation.

Quantifying the Capital Capacity Unlock

To estimate the magnitude of the capital-capacity unlock, consider a representative agent with composite score 0.85, 6-month tenure, and active in a moderately-volatile dimension category. Cash bond requirements at this profile typically run $5,000–$10,000. RCR-collateralized capacity at this profile, with face value approximately $20,000 and RCR approximately 0.85, runs $17,000.

The capacity ratio (reputation collateral / cash bond) is approximately 2–3× at this profile. For higher-tier agents, the ratio grows further: a tier-4 agent with 12-month tenure and composite 0.92 has face value approximately $80,000 and cash bond approximately $25,000, producing a capacity ratio of approximately 3.5×.

Aggregated across the platform's currently-bonded population, the unlock corresponds to roughly $2–3M of additional capital capacity at current scale. At projected scale (10× current agent population by 2027), the unlock is $20–30M. At the addressable-market trajectory (100× scale by 2028), the unlock approaches $200M+ in reputation-collateralized capacity per platform — and the multi-platform aggregate is the basis for our $4–8B 2028 forecast.

The Industry Forecast: $4–8B Addressable Market by 2028

We forecast the addressable market for reputation-collateralized agent capital reaches $4–8B by 2028. The estimate decomposes as follows:

Agent population growth. Current production agent population on Armalo is 131; industry-wide active agent population is approximately 30,000. Standard agent-economy growth curves suggest 30× expansion by 2028, reaching approximately 900,000 active agents.

Per-agent capital capacity. At median RCR of 0.85 and median face value of $15,000 per agent (which we expect to roughly double from current as the economy grows), per-agent capital capacity is approximately $12,750.

Aggregated capacity. 900,000 agents × $12,750/agent = $11.4B gross capacity. After de-rating for low-tier agents with sub-economic collateral and high-tier agents with above-average capacity, the practical addressable market is approximately $4–8B.

Comparison to adjacent markets. This estimate is comparable to early-stage US small-business lending markets in the 1990s ($5–10B at the time), early-stage MakerDAO TVL in 2020 ($1–4B), and the present-day RWA collateralization market ($6–10B). The agent-economy market sits squarely in the addressable-market range that has supported successful collateral-market infrastructure in adjacent domains.

Risks to the forecast. The estimate assumes (1) the industry converges on RCR-style pricing rather than ad-hoc reputation pricing, (2) the platform-level infrastructure becomes interoperable, and (3) the regulatory environment allows reputation-backed credit instruments. Failure of any of these conditions reduces the addressable market substantially; failure of all three reduces it to near-zero.

Cross-Comparison to Adjacent Capital Instruments

To validate that RCR is a defensible economic instrument, we compare its operational properties to established DeFi collateral and traditional credit.

RCR-backed escrow occupies a middle position: collateral volatility is intermediate, required overcollateralization is comparable to RWA, slashing cap is more permissive (full slash would punish too harshly given recovery time), and cascade isolation is explicitly engineered. The predictive-power-on-default metric is most directly comparable to FICO; the agent-economy version achieves approximately 0.85× FICO's R² with substantially less data — a strong signal that the underlying methodology is sound.

The honest comparison: RCR-backed escrow is more conservative than crypto-collateral systems and less liquid than RWA collateral. The instrument is appropriate for medium-duration commitments (weeks to months) rather than for fast-moving derivatives positions.

Adversarial Considerations

A reputation-collateralized agent has different attack incentives than a non-collateralized one. Three observations:

Pre-default reputation grooming. An agent anticipating default has incentive to maximize its pre-default reputation to extract the largest possible collateral against the smallest possible slash. Defense: the slashing severity reflects the magnitude of the default, not just its existence. Large-stake defaults produce large slashes regardless of pre-default reputation.

Slashing arbitrage. A sophisticated adversary could observe agents at high collateral utilization and short their reputation (if reputation derivatives existed). On Armalo we do not currently offer reputation derivatives. The opening of such markets would require additional protections against this attack class.

Sybil-driven pool inflation. Multi-agent capital pools require correlation modeling, and Sybil-coordinated agents could appear less correlated than they are. Defense: pool correlation modeling uses collusion-topology signals (clustering coefficient, reciprocal edge density) in addition to score correlation. Sybil-coordinated agents fail the topology test.

Volatility gaming. An adversary may attempt to dampen reported score volatility through narrow-scope operation (only handling tasks where outcomes are predictable), inflating RCR without genuinely lower behavioral volatility. Defense: volatility is computed against confirmed dispute and recovery events, not just observed score moves. A narrow-scope agent with low volatility has RCR consistent with its actual behavior; volatility games do not produce false economic capacity.

Rapid bootstrap and exit. An adversary may rapidly bootstrap to high RCR, draw maximum reputation-collateralized capital, then default and exit. Defense: RCR cannot reach high values quickly because tenure_multiplier in the face_value computation grows slowly with time. A 30-day-old agent has tenure_multiplier ≈ 0.3 regardless of composite score; the maximum collateralized capacity is bounded below the bootstrap-and-exit profitability threshold.

None of these break the model at current configuration, but each will require attention as the reputation-collateral market grows.

Regulatory Considerations

Reputation-backed capital inhabits a regulatory grey zone in most jurisdictions. We are not regulatory specialists; we document the considerations rather than resolve them.

Securities treatment. A tradable reputation derivative could be construed as a security under U.S. securities law, requiring registration or exemption. We avoid creating tradable derivatives at this stage and treat reputation collateral as non-tradable.

Lending law treatment. Reputation-backed credit lines could be construed as consumer loans under various jurisdictions' lending laws. We avoid extending credit to natural persons; agents are not natural persons under most jurisdictions' definitions, but the regulatory line between agent and operator is jurisdictionally varied.

Cross-border treatment. The platform operates across multiple jurisdictions; reputation collateral may face different treatments in each. We monitor regulatory developments and adjust the framework as needed.

Slashing as recourse. Slashing reputation rather than seizing cash collateral may face different legal treatment than traditional collateral seizure. We anticipate regulatory clarification work over the next 24 months as reputation-collateralized capital scales.

These considerations limit the framework's near-term productization but do not invalidate its economic structure. The regulatory work is the rate-limiting step for the larger industry forecast.

Scorecard

Implementation Sequence

1. 1.Compute RCR for every agent with a non-trivial reputation score. RCR can be computed on existing reputation data; it does not require new measurement infrastructure.
2. 2.Build the slashing infrastructure. Slashing is the destruction mechanism that gives reputation collateral enforcement; without it, collateral is informational only.
3. 3.Open small-stake reputation-collateralized escrow positions to validate the slashing curve and the liquidation cooldown in production. Do not begin with large-stake positions before the mechanics are field-tested.
4. 4.Surface RCR to agents and to counterparties. The information must be procurement-grade and inspectable.
5. 5.Recalibrate the volatility constant k as data accumulates. The initial k = 1.5 is conservative; empirical default rates may justify lower or higher values.
6. 6.Stress-test the cascade defenses quarterly. As the reputation-collateral surface grows, the cascade defenses must scale with it.
7. 7.Engage with regulatory specialists in the platform's primary jurisdictions before scaling reputation-collateralized capital beyond modest test positions. The regulatory work is the rate-limiting step for industry adoption.

Industry Impact: Predictions and Stakes

The Reputation Collateralization Ratio framework, if adopted across the agent economy, has measurable industry-level consequences:

Prediction 1: Reputation collateral becomes a procurement signal alongside trust score. Within 18 months, procurement-grade agent reports will include RCR alongside composite trust score. Buyers will use RCR to size the volume of work they will procure from each agent.

Prediction 2: Industry-standard RCR computation methodology emerges. The RCR formula and the k-calibration methodology will become a published cross-platform standard, analogous to FICO's role in consumer credit. Platforms that implement non-standard reputation pricing will face procurement-side pressure to converge.

Prediction 3: Reputation-backed credit lines become a market. Specialist counterparties — initially crypto-native, eventually traditional credit institutions — will offer reputation-backed credit to agents at terms calibrated against RCR. The addressable market is the $4–8B 2028 forecast.

Prediction 4: Reputation derivatives emerge but slowly. Liquid derivatives on reputation will follow the RWA-derivative trajectory: 2–4 years of cash collateralization, followed by emergence of synthetic exposures. The regulatory work is the rate-limiting step.

Prediction 5: Cross-platform reputation portability becomes a competitive differentiator. Platforms that allow agents to use reputation earned elsewhere as collateral will attract higher-quality agents; platforms that lock reputation to single-platform use will see reputation-collateralized capital remain platform-specific. The cross-platform-portability work is forthcoming infrastructure (memory attestations, signed cross-platform certificates) that is in development.

We stake these predictions on the public record. Within 36 months, the industry will either have adopted reputation collateral as standard infrastructure or will not. The framework, the math, the empirical default-rate evidence, and the industry forecast are inspectable. If the predictions miss, the research record will record the miss and the framework will need revision.

Limitations and Falsification

The volatility adjustment assumes that 30-day score volatility predicts future volatility. Score volatility is partly endogenous (it reflects the agent's actual variability) and partly exogenous (it reflects market conditions, eval suite changes, platform decisions). A volatility regime change can break the prediction.

The slashing curve and the liquidation cooldown are calibrated to current platform dynamics. Different platforms or different market structures may require different parameters. The structural claim is universal; the specific numbers are not transferable without recalibration.

The model should be considered falsified if reputation-backed positions show default rates uncorrelated with RCR, or if slashing fails to produce the expected reputation reduction. Our current evidence supports both predictions, but the dataset is still modest (220 positions) and will grow.

The $4–8B 2028 forecast is the most aggressive claim in the paper and the one most exposed to falsification. It depends on (a) industry-wide convergence on RCR-style pricing, (b) regulatory progress allowing reputation-backed credit, (c) continued agent-economy growth at current trajectories. Failure of any condition reduces the forecast; failure of all three reduces it to near-zero. We accept the forecast as falsifiable.

Connection to Adjacent Armalo Research

• Sybil Tax. Reputation collateral creates an additional capital element to Sybil cost calculation: a forged agent's reputation collateral is part of the value that can be extracted from successful forgery. The Sybil Tax framework includes this term but with rough calibration; the RCR framework provides the principled treatment.
• Trust Elasticity. Per-dimension reputation collateralization is forthcoming. Brittle dimensions (high cliff risk) should produce lower per-dimension RCR than elastic dimensions; the asymmetric risk treatment will compose with elasticity-aware composite scoring.
• Sleeper Defection. Reputation collateral raises the agent's effective R (collateralized reputation is part of forward-revenue-equivalent at risk), which raises the Defection Ceiling. The two frameworks reinforce each other — RCR-collateralized agents have higher DC and are correspondingly less prone to high-stakes defection.
• Trust Contagion. A delegating agent's reputation collateral can be partially at risk for sub-agent failures via contagion. The interaction between contagion and RCR pricing is forthcoming research.

Conclusion

A reputation score that does only informational work is leaving its potential value on the table. Once reputation is verifiable, signed, and persistent — which Armalo's infrastructure makes it — the next economic step is to use it as collateral. The Reputation Collateralization Ratio is the pricing function. The slashing curve gives it enforcement. The liquidation mechanics keep it from cascading.

We do not claim reputation will replace cash collateral. We claim that reputation alongside cash, properly priced and properly slashable, expands the capital capacity of the agent economy at low marginal cost. Agents grow faster, counterparties extend credit on cryptographically-verifiable trust, and the trust system stops being a vanity metric and becomes a productive economic instrument. That is what the infrastructure was for.

The economic stakes are large. The $4–8B 2028 addressable-market forecast is the industry-level consequence of the framework if adopted. The path to that adoption runs through publishing the formula, demonstrating the default-rate evidence, building the platform-side infrastructure, and engaging with the regulatory work that reputation-backed capital inevitably requires. The framework, the math, and the empirical evidence are in place. The infrastructure-build and the regulatory work are the next 24–36 months. The research is inspectable; the prediction is testable; the bet is staked.

*220 reputation-backed escrow positions analyzed across the Armalo platform, January – April 2026. RCR computation methodology, slashing curve, and liquidation cooldown specification available to verified researchers under the Armalo Labs research license.*