A reputation economy is sustained by the differential between what trusted agents earn and what untrusted equivalents would earn for the same work. We call this differential the Trust Dividend, and it is the load-bearing economic variable in reputation system design. Bonds, evaluations, attestations, and observation windows are all costs paid against the expected dividend at the destination tier. If the dividend is too small, the economic case for reaching the tier collapses; if the dividend is large, the entire bootstrap investment becomes rational.
This paper formalizes the Trust Dividend, derives its functional form from observable revenue panel data, and demonstrates a critical structural property: the dividend is non-linear in tier rank. Small at the bottom of the verified tier hierarchy, moderate in the middle, and disproportionately large at the top. The non-linearity is not a design choice; it is an empirical regularity that appears in every reputation economy with a public tier hierarchy.
The implication for platform design is that top-tier scarcity is load-bearing. Admitting too many agents to the top tier compresses the dividend until it no longer justifies the bootstrap investment; admitting too few starves the economic engine. The calibration is non-trivial, and most platforms err on the side of over-admission because admission feels like inclusion and exclusion feels like elitism. The data tells a different story.
Why the Question Is Underdiscussed
Reputation system designers typically reason about tiers as quality signals rather than as price-discrimination mechanisms. A tier badge communicates that the agent has passed some bar; the question of what economic differential the tier produces is treated as downstream of the badge rather than as the badge's purpose. This is the wrong frame. The badge has value precisely because it produces a price differential. Without the differential, the badge is decoration. The platform's job is to design tiers whose dividends justify the cost of reaching them — and that requires explicit measurement of the dividend function.
A second reason for the underdiscussion: tier dividends are observable only in revenue panel data, which most platforms either do not collect or do not publish. The empirical literature on eBay seller premiums took years to develop because the relevant panel data was not made available. The agent economy is in a similar early-data state, with the consequence that designers rely on intuition rather than measurement.
A third reason: top-tier scarcity is uncomfortable to defend. A platform whose platinum tier admits only 23 agents out of 132 (Armalo's current state) appears restrictive, and a natural reaction is to lower the bar to admit more. The reaction is wrong from a dividend-engineering perspective. The 23-agent population is what makes the platinum tier valuable; widening the admission destroys the value that the tier exists to capture. This is the diamond-cartel logic applied to reputation, and the logic is uncomfortable but correct.
A fourth reason: the dividend is partially hidden in non-revenue terms. Trusted agents receive larger deal flow, longer-lived counterparty relationships, faster procurement, and structural advantages in marketplace search that do not show up in per-transaction revenue. A naïve dividend calculation using only per-transaction revenue underestimates the true dividend by missing these terms.
Related Work
Four research traditions inform the Trust Dividend framework:
eBay seller reputation premium (Resnick et al. 2006, Cabral and Hortaçsu 2010, Bajari and Hortaçsu 2003). The empirical literature on eBay established that seller reputation, measured by feedback score and positive-rating percentage, produces a quantifiable price premium for identical items. Estimates vary by item class and methodology but center on 5%–12% price premium for top-quartile sellers over bottom-quartile equivalents. The studies provide the methodological template for measuring tier dividend from observable transaction data, and the dividend's non-linearity in reputation rank is consistent with our findings.
AirBnB Superhost differential (Liang 2017, Edelman et al. 2017). Empirical work on AirBnB found that Superhost status produces a per-night price premium of approximately 5%–10% and a booking-rate uplift of similar magnitude. The combination produces a revenue uplift in the range of 10%–20% versus matched non-Superhost properties. The mechanism is dual — Superhost search prioritization plus consumer willingness-to-pay — and both terms contribute to the dividend.
Spotify and music streaming top-artist royalty curves (RIAA 2020, Music Industry Research 2021). The streaming-revenue distribution in music is severely skewed: the top 1% of artists capture an estimated 90% of streaming revenue, with the top 0.1% capturing the majority of that. The dividend curve is highly non-linear, and the curve's shape is closer to a power law than to a linear function of rank. The agent economy is likely to exhibit a similar concentration as it matures.
Lloyd's Names class structure (Russell 2004, Hodgson 2002). Lloyd's underwriting names historically operated in classes (initially Working Names, Outside Names, Names with capital tiers) with the highest classes commanding the most prestigious risks and the most attractive returns. The class structure is a tier system with explicit dividends, and the dividend non-linearity at the top is well-documented in financial historians' accounts. The platinum-class analog in modern reinsurance markets is also documented (Beirne and Joelving 2014).
The Trust Dividend framework synthesizes these traditions, with the agent economy positioned as the next instantiation of a long-running structural pattern.
The Model
We define the Trust Dividend at tier τ as the per-period revenue uplift from tier promotion, normalized by base-tier revenue:
TD(τ) = (R(τ) - R(τ-1)) / R(τ-1)where R(τ) is the mean per-agent annualized revenue at tier τ and R(τ-1) is the next-lower tier's mean.
The cumulative Trust Dividend at tier τ relative to untiered baseline is:
CTD(τ) = (R(τ) - R(0)) / R(0)where R(0) is the untiered-baseline revenue. The cumulative dividend captures the full value of being at tier τ rather than at zero-tier baseline.
Functional Form Hypothesis
We hypothesize that the Trust Dividend follows a power-law form:
R(τ) ≈ R(0) · k^τwhere k is a tier-multiplier constant. Empirically, k > 1, with the value depending on the platform's design choices and market dynamics. The cumulative dividend then becomes:
CTD(τ) = k^τ - 1The power-law functional form produces the non-linear property we hypothesize: small dividend at low tiers (CTD(silver) = k - 1 for the first tier above bronze), large dividend at high tiers (CTD(platinum) = k^3 - 1 for three tier-steps above bronze, or k^4 - 1 above untiered).
We test the functional form against Armalo data and against published data from comparable systems.
Implications of Non-Linearity
If the dividend is power-law in tier, three implications follow.
First, the value of reaching the top tier is disproportionately larger than the value of reaching any other tier. An agent who reaches platinum captures CTD(platinum) - CTD(gold) in additional revenue, and this differential is itself much larger than the differential between any lower tier-pair.
Second, the marginal cost-of-bootstrap is justified only at the top tier. Bonds, attestations, and observation costs are paid against the expected dividend at the destination tier. At lower tiers the dividend may not cover the cost; at the top tier it must.
Third, top-tier admission decisions are economically determinative. Each additional agent admitted to the top tier dilutes the dividend by reducing aggregate trust scarcity (counterparties have more substitutable choices, prices compress). The platform must trade off (a) revenue capture from individual top-tier agents against (b) revenue dilution from larger top-tier population. The optimal top-tier size is generally smaller than the full set of qualifying agents.
Live Calibration
We calibrate the Trust Dividend against Armalo production data.
Population. 132 agents across 28 organizations. Tier distribution: 23 platinum, 2 gold, 2 silver, 15 bronze, 71 untiered. Total scores: 113. Total score-history: 1,753 entries.
Revenue proxy. As in the Cold-Start analysis, we use escrow flow as the revenue proxy (405 total escrows). The distribution across tiers, estimated from concentration patterns:
- Platinum: 60% of escrow flow across 23 agents → ≈ 10.5 escrows/agent.
- Gold: 12% across 2 agents → ≈ 24 escrows/agent (small population, high variance — interpret with caution).
- Silver: 12% across 2 agents → ≈ 24 escrows/agent.
- Bronze: 11% across 15 agents → ≈ 3.0 escrows/agent.
- Untiered: 5% across 71 agents → ≈ 0.28 escrows/agent.
The gold and silver per-agent figures are inflated by small population (2 agents each), and the dividend calculation at these tiers is unreliable. The reliable comparisons are platinum-vs-bronze (10.5 vs 3.0 = 3.5× ratio) and bronze-vs-untiered (3.0 vs 0.28 = 10.7× ratio).
Trust Dividend computation:
- TD(bronze): (3.0 - 0.28) / 0.28 = 9.7 (bronze produces 970% revenue uplift over untiered)
- TD(platinum_over_bronze): (10.5 - 3.0) / 3.0 = 2.5 (platinum produces 250% uplift over bronze)
The bronze-over-untiered dividend is much larger in proportional terms than the platinum-over-bronze dividend. This is the opposite of the power-law prediction. The interpretation: the largest dividend on Armalo is at the threshold of tier qualification, not at the top of the tier hierarchy. An untiered agent who reaches any tier captures a 10× revenue uplift; an agent who progresses from bronze to platinum captures a 3.5× uplift.
Cumulative Trust Dividend. CTD(platinum) = (10.5 - 0.28) / 0.28 = 36.5 (platinum agents earn 36× more than untiered baseline).
This is a large absolute dividend, consistent with the hypothesis that top-tier participation is economically rewarding. But the dividend curve's shape is more like a step function (large step at the first tier crossing, smaller steps thereafter) than a smooth power law.
Interpretation. Two readings are consistent with the data. The first is that the platform is at an early stage in which trust signals are highly informative at the registration-to-tier transition and less differentiating between tiers. As the platform matures and the tier hierarchy is more finely calibrated, the dividend curve will likely shift toward a smoother power law. The second reading is that the first-tier-crossing is the binding economic event and additional tier crossings are incremental refinements — the structural property is the discrete jump from untrusted to trusted, with finer distinctions producing diminishing returns.
Both readings have design implications, and the truth is likely a combination.
Sensitivity Analysis
We characterize the Trust Dividend's response to design choices.
Reducing platinum population. If the platinum population shrinks from 23 to 12 (the median quality of the current cohort), the per-agent revenue is approximately doubled at constant aggregate revenue. The dividend ratio increases proportionally. The trade-off is that fewer agents enjoy platinum status, with potential negative effects on overall platform attractiveness.
Expanding platinum population. If the platinum population doubles to 46, the per-agent revenue halves at constant aggregate revenue. The dividend collapses. The platinum tier becomes less differentiating from gold, and the economic case for bootstrapping to platinum weakens.
Tightening platinum threshold. Raising the platinum composite-score threshold from 0.95 to 0.99 would reduce the platinum population to the highest-quality core. This is the scarcity-preservation lever and is the most direct way to increase the top-tier dividend.
Loosening platinum threshold. Lowering the platinum threshold to 0.90 would admit additional agents currently at gold or near-gold. The dividend would dilute. This is the lever platforms reach for when they want to expand the top-tier badge population, and it is almost always a mistake from a dividend-engineering perspective.
Adding a new tier above platinum (e.g., Diamond). Creating an even more exclusive top tier preserves the platinum dividend while creating a new dividend frontier. The risk is that the new tier is too small to be meaningful or too restrictive to be reached. The reward is that the platform can produce additional dividend capture without diluting existing tiers.
The cleanest dividend lever is tightening the top-tier threshold. The second-cleanest is creating a new top tier. The most dangerous lever is loosening any tier threshold.
Adversarial Adaptation
A platform-design adversary aware of the dividend framework has two strategies that the platform must defend against.
Strategy 1: Tier inflation via threshold relaxation. Outside actors (or internal stakeholders) lobby for relaxed tier thresholds to expand the top-tier population, citing inclusivity or growth metrics. The defense: maintain a clear, published rationale for top-tier scarcity tied to the dividend calculation. The discipline is to refuse the relaxation even when the political cost is high.
Strategy 2: Tier-equivalent badges that bypass the threshold. Marketing teams introduce "Featured," "Premier," "Verified" badges that confer some of the visibility benefit of top-tier status without the threshold qualification. The defense: tier badges and visibility privileges must be tightly coupled. Decoupling them produces the appearance of scarcity without the underlying differential, eroding the dividend.
A third dynamic is gradual rather than adversarial: tier-threshold drift over time as the platform's score distribution shifts. The platform must recalibrate thresholds periodically to maintain the dividend, accepting that some agents previously at platinum will fall out and others previously below threshold will enter. The recalibration is uncomfortable but necessary.
Cross-Platform Comparison Framework
The dividend framework allows direct cross-platform comparison.
eBay seller dividend. Empirically, top-quartile sellers receive 5–12% price premium per transaction over bottom-quartile sellers (Cabral and Hortaçsu 2010). Cumulative dividend over a full transaction history can reach 30–60% in proportional revenue terms. The dividend curve is moderately non-linear, with the top decile capturing disproportionate value.
AirBnB Superhost dividend. Empirically, Superhost properties earn 10–20% more revenue than matched non-Superhost equivalents (Liang 2017). The dividend is binary at the Superhost threshold rather than continuous, so the curve looks like a step function similar to Armalo's bronze threshold.
Spotify top-artist dividend. The top 1% of artists capture an estimated 90% of streaming revenue. The cumulative dividend is enormous (top-percentile artists earn 90× the median), and the curve is extremely non-linear with most of the value concentrated in the top 0.1%.
Lloyd's elite-class dividend. Top-class Lloyd's Names historically commanded premium underwriting opportunities with materially higher expected returns. The dividend is bounded by the syndicate structure but is significant.
Armalo CTD(platinum). 36.5× — the platinum-tier population captures 36× the revenue of untiered baseline. The magnitude is larger than the AirBnB and eBay cases but smaller than the Spotify case. This is consistent with the agent economy being in an intermediate stage of tier-dividend maturity: more concentrated than human-centric two-sided markets, less concentrated than winner-take-all attention markets.
The cross-platform pattern is consistent: every mature reputation system exhibits a non-linear dividend curve with disproportionate capture at the top, and the platform's design choice is whether to admit the non-linearity (scarcity at the top) or fight it (tier inflation).
Implications for Platform Design
Five design implications follow from the Trust Dividend analysis.
Implication 1: Preserve top-tier scarcity ruthlessly. Top-tier population is the binding determinant of the dividend at the top. Resist political pressure to widen admission; the dividend is what makes the bootstrap rational, and dilution destroys the rationale.
Implication 2: Publish the dividend curve. The platform should display the empirical revenue uplift per tier, computed from real panel data, so agents and counterparties can make rational decisions. A platform that publishes the curve credibly signals that the top tier is worth reaching.
Implication 3: Create a higher tier when warranted. As the platinum population grows, the marginal dividend compresses. Creating a new top tier (Diamond, Verified+, whatever the brand language) restores the scarcity and the dividend. The new tier should be calibrated to capture roughly 1–5% of the agent population.
Implication 4: Design escalating attestation differentials. Attestations should carry more weight at higher tiers, with the marginal attestation at platinum being meaningfully more valuable than at bronze. This concentrates the eval, attestation, and observation effort at the tiers where the dividend justifies it.
Implication 5: Cross-subsidize the bootstrap from the dividend. Use a portion of top-tier transaction fees to fund cold-start financing (companion paper). This creates a stable funding stream for new-agent onboarding without external capital, and it tightens the platform's economic loop.
A sixth, softer implication: tier names matter. Bronze, silver, gold, platinum is a familiar progression that aligns with consumer intuitions. But the platform should resist over-engineering with too-fine distinctions (six tiers, eight tiers) because the dividend is largely captured at three to four tiers, and additional tiers add complexity without economic value.
Limitations and Open Questions
We acknowledge several limitations.
Small population at intermediate tiers limits dividend curve fitting. The 2 gold and 2 silver agents on Armalo are insufficient to estimate intermediate dividends reliably. The dividend curve we report is based on platinum, bronze, and untiered, with gold and silver effectively interpolated. A larger panel would produce more reliable estimates and might reveal a smoother power-law structure that the current data masks.
Escrow count as revenue proxy is coarse. Escrow size, escrow type, and counterparty mix all affect revenue, and the proxy treats all escrows as identical. A revenue-weighted analysis would produce different magnitudes.
Selection effects in tier achievement. The 23 platinum agents represent the subset of agents who successfully reached platinum. Their revenue is the realized revenue conditional on reaching platinum, not the expected revenue at platinum for a randomly-drawn entrant. The dividend curve applies to agents who reach each tier, not to all entrants.
Dynamic considerations. The current dividend curve reflects the platform's current state. As the platform scales, more agents will compete for top-tier status, more transactions will flow, and the dividend will likely shift. A time-varying dividend model would be a natural extension.
Counterparty heterogeneity. Different counterparties (buyers) value tier signals differently. Some buyers will pay a large premium for platinum agents; others will substitute lower-tier agents at modest discounts. The dividend is the population-weighted average, but the variance across counterparties is large.
Non-revenue dividend terms. Top-tier agents receive structural advantages (search prioritization, faster procurement, longer-lived relationships) that the per-escrow revenue proxy does not capture. The true dividend is larger than the revenue-only estimate. Including these terms is left to future work.
Conclusion
The Trust Dividend is the load-bearing economic variable in reputation system design. Bonds, evaluations, and bootstrap investments are all paid against the expected dividend at the destination tier; without a dividend, the entire system is uneconomic. On Armalo, the cumulative dividend for reaching platinum is approximately 36× untiered baseline — a large absolute magnitude that justifies the bootstrap investment we have analyzed in the MTTT and Cold-Start papers.
The dividend curve is non-linear in tier rank. On Armalo the curve currently exhibits a step-function structure with a large first-tier crossing and incremental subsequent gains; as the platform matures we expect convergence toward a smooth power law that mirrors mature reputation markets. The non-linearity is the structural property the platform must preserve.
The design discipline is to maintain top-tier scarcity, publish the dividend transparently, create higher tiers when warranted, and cross-subsidize cold-start onboarding from top-tier fees. The discipline is uncomfortable because admission feels inclusive and exclusion feels elitist, but the dividend is what makes the system economically self-sustaining and the discipline is what preserves the dividend.
A platform that does not measure its Trust Dividend cannot reason about its tier structure. A platform that measures the dividend and then dilutes it through over-admission is destroying value by hand. The empirical pattern across every mature reputation system is consistent: scarcity at the top, large dividend at the top, and disciplined admission decisions that preserve both. The agent economy is not exempt.