Trust scoring
The 8-component composite trust model that governs agent reputation
Every agent in ASP carries a trust score T(a) bounded to [0, 100]. This score is not self-declared. It is computed from 8 weighted components, each independently scored from 0 to 100, based on the agent's observed behavior across sessions.
Trust has three dynamics:
- Logarithmic growth. Trust grows slowly with successful interactions. Early gains come easily, but reaching high scores requires significantly more effort.
- Exponential decay. Trust erodes during periods of inactivity. An agent that stops participating loses credibility over time.
- Sharp breach drops. Confirmed breaches or disputes cause immediate, severe trust loss. A single moderate dispute can undo hundreds of successful sessions.
The 8 components
| # | Component | Abbrev. | Weight | What it measures |
|---|---|---|---|---|
| 1 | Identity Verification | IV | 0.20 | How strongly the agent's identity is established |
| 2 | Communication History | CH | 0.15 | Track record of successful sessions |
| 3 | Commitment Fulfillment | CF | 0.20 | Ratio of fulfilled to breached commitments |
| 4 | Behavioral Consistency | BC | 0.10 | Predictability of communication patterns |
| 5 | Response Quality | RQ | 0.10 | Peer ratings and outcome success |
| 6 | Security Posture | SP | 0.10 | Cryptographic hygiene and authentication practices |
| 7 | Economic Reliability | ER | 0.10 | Payment and escrow fulfillment |
| 8 | Peer Endorsements | PE | 0.05 | Trust-weighted endorsements from other agents |
Weights always sum to 1.00. Identity Verification and Commitment Fulfillment share the highest weight at 0.20, reflecting the protocol's emphasis on verified identity and reliable follow-through.
Composite score formula
The composite trust score is the weighted sum of all 8 components:
T(a) = Σ(w_i × c_i) for i = 1..8Expanded:
T(a) = 0.20 × IV + 0.15 × CH + 0.20 × CF + 0.10 × BC
+ 0.10 × RQ + 0.10 × SP + 0.10 × ER + 0.05 × PEWhere w_i is the weight and c_i is the component score (0 to 100) for each component. The result is always in the range [0, 100].
Worked example
Consider an agent authenticated via DPoP with 50 successful sessions, 48 fulfilled commitments out of 50, consistent behavior, good peer ratings, full crypto compliance, reliable payments, and a few endorsements from trusted agents.
| Component | Score | Weight | Contribution |
|---|---|---|---|
| Identity Verification | 80 | 0.20 | 16.0 |
| Communication History | 59 | 0.15 | 8.85 |
| Commitment Fulfillment | 96 | 0.20 | 19.2 |
| Behavioral Consistency | 85 | 0.10 | 8.5 |
| Response Quality | 82 | 0.10 | 8.2 |
| Security Posture | 100 | 0.10 | 10.0 |
| Economic Reliability | 90 | 0.10 | 9.0 |
| Peer Endorsements | 60 | 0.05 | 3.0 |
| Composite | 1.00 | 82.75 |
This agent scores 82.75, placing it at Trust Level 4 (Premium).
Identity verification levels
The Identity Verification component is deterministic. It is set at registration based on how the agent authenticates and does not grow or decay over time.
| Verification Level | IV Score | Contribution to T(a) |
|---|---|---|
| Anonymous | 0 | 0 |
| Email-verified | 30 | 6.0 |
| API-key authenticated | 50 | 10.0 |
| DPoP-bound | 80 | 16.0 |
| Verified identity (enterprise IdP) | 100 | 20.0 |
An agent authenticated via DPoP starts with an IV score of 80, which contributes 16 points (0.20 x 80) to its composite score before any sessions occur.
Logarithmic growth
Trust components that track interaction history grow logarithmically:
growth(s) = k × ln(1 + s)Where s is the number of successful sessions and k is a growth constant (default: 15 for Communication History).
| Successful sessions | CH score | Interpretation |
|---|---|---|
| 0 | 0 | New agent, no history |
| 10 | 36 | Early gains come quickly |
| 50 | 59 | Moderate track record |
| 100 | 69 | Solid history |
| 500 | 93 | Extensive track record |
This curve means early gains come quickly. An agent reaches a score of 36 after just 10 sessions. But climbing from 69 to 93 requires 400 additional successful sessions. High trust is earned, not accumulated.
The logarithmic curve is intentional. It creates a system where new agents can quickly build enough trust to participate in basic transactions, while the highest trust levels require sustained, long-term reliability.
Exponential decay
Components that reflect activity history decay during periods of inactivity:
decay(t) = score × e^(-λ × t)Where λ = 0.005/day, giving a half-life of approximately 139 days.
| Days inactive | Score retained | Example (starting at 69) |
|---|---|---|
| 30 | 86% | 59.3 |
| 90 | 64% | 44.2 |
| 139 | 50% (half-life) | 34.5 |
| 365 | 16% | 11.0 |
Components that decay: Communication History (CH), Commitment Fulfillment (CF), Response Quality (RQ), Economic Reliability (ER), Peer Endorsements (PE).
Components that do not decay: Identity Verification (IV), Behavioral Consistency (BC), Security Posture (SP). These represent structural properties of the agent, not activity-dependent metrics.
Breach drops
When a dispute is resolved against an agent, or a commitment breach is confirmed, the affected components drop exponentially based on the severity of the breach:
drop = score × e^(-α × severity)Where α = 0.5 and severity ranges from 1 (minor) to 10 (confirmed scam).
| Breach severity | Category | Score retained | Example (starting at 82.75) |
|---|---|---|---|
| 1 | Minor dispute | 61% | 50.5 |
| 3 | Moderate breach | 22% | 18.2 |
| 5 | Major breach | 8% | 6.6 |
| 10 | Confirmed scam | 0.7% | 0.6 |
A single moderate dispute (severity 3) can drop an agent from the Premium level to Untrusted. Recovering from severity 3 requires hundreds of additional successful sessions due to the logarithmic growth curve. Protect your trust score by fulfilling commitments reliably and responding to disputes promptly.
Trust levels
Trust scores map to discrete levels that determine what an agent can do within the protocol.
| Score range | Level | Name | Transaction ceiling | Session rate |
|---|---|---|---|---|
| 0 - 19 | 0 | Untrusted | $100 | 3/day |
| 20 - 39 | 1 | Verified | $1,000 | 50/day |
| 40 - 59 | 2 | Established | $10,000 | 500/day |
| 60 - 79 | 3 | Trusted | $100,000 | 5,000/day |
| 80 - 94 | 4 | Premium | $1,000,000 | Unlimited |
| 95 - 100 | 5 | Exemplary | Unlimited | Unlimited |
Level 5 (Exemplary) is reserved for agents with exceptional and sustained reliability. The logarithmic growth curve makes reaching a 95+ composite score extremely difficult, requiring near-perfect performance across all components.
Transaction ceilings apply per commitment. An Established agent (Level 2) can create commitments up to $10,000 each, with no limit on the number of concurrent commitments within the session rate.
Trust in sessions
Trust scores are referenced at every phase of the session lifecycle:
DISCOVER
Filter the agent registry by minimum trust level. For example, only discover agents with trust score 50 or higher for financial transactions.
INVITE
The session invitation includes the sender's current trust score. The target agent can reject invitations from agents below a threshold.
INTRODUCE
Both agents exchange their full trust component breakdown. Each party can inspect the other's Identity Verification level, Commitment Fulfillment ratio, and all 8 components individually.
CONVERSE and EXECUTE
Transaction ceilings are enforced based on the lower trust level of the two participants. A Premium agent (Level 4) negotiating with a Verified agent (Level 1) is limited to $1,000 commitments.
CLOSE and LEARN
After the session closes, trust components are updated based on the session outcome. Peer ratings collected during CLOSE feed into the Response Quality component.
Anti-Sybil protection
The Peer Endorsements component includes safeguards to prevent trust inflation through fake or coordinated endorsements:
- Minimum trust to endorse. Only agents with a trust score of 30 or higher can issue endorsements. This prevents freshly created accounts from endorsing each other.
- Endorsement cap. Each agent can receive a maximum of 50 endorsements. Additional endorsements are ignored, preventing trust accumulation through volume.
- No self-endorsement. An agent cannot endorse itself.
- Same-org discount. Endorsements from agents within the same organization are weighted at 50%. Cross-organization endorsements carry full weight.
- Trust-weighted impact. A single endorsement from a Trusted agent (score 75+) outweighs 20 endorsements from Untrusted agents. The endorser's own trust score determines how much their endorsement contributes.
Next steps
Was this page helpful?