Source: https://trusteed.xyz/en/trust/methodology

Trust score · methodology

# How Trust Score v4.1 is calculated

Trust Score v4.1 combines three weighted dimensions with an optional external-assurance bonus and confidence-tier caps. Each response's \`engines\` field names the engine actually used: v4.1 for both overall and breakdown, or v2 for both. Two different situations fall to v2 — insufficient\_data (no applicable evidence for that store, a normal business outcome) and v2\_degraded (the overlay is off or failed technically) — read \`confidence\_level\` to tell them apart. A legacy v2 component breakdown does not explain an overall score produced by v4.1.

Methodology snapshot reviewed 2026-09-13. Check the live endpoint for current weights and engine selection.

[View the live signed methodology response ↗](https://trusteed.xyz/api/v1/trust/methodology)[Back to Trust overview](https://trusteed.xyz/trust)

Current methodology · per-response engine 

## v4.1: three weighted dimensions and an optional bonus

Read engines.overall and engines.breakdown before interpreting a score. overall uses 0–1; trust\_score\_v41 uses 0–100 when available — the same score on two scales (0.62 and 62), never two different numbers to average. In this deployment (coherent engine on), engines.mixed is false: overall and breakdown agree on the same engine — with insufficient\_data (no applicable evidence) or v2\_degraded (overlay off or technically failed), both are v2; in any other case, both are v4.1. If the coherent condition were off, overall could come from v4.1 while breakdown stayed v2 (mixed: true); always read both fields. insufficient\_data is missing usable evidence, not a measured zero; the public route can return a degraded v2 fallback.

v4.1 available · v2 fallback possible · read engines

### Three weighted dimensions + External Assurance

v4.1 weights depend on confidence and are renormalized over applicable dimensions. External Assurance adds up to 10 points after the base score, subject to its gates and the tier cap; it is not a fourth weighted dimension.

**Merchant reliability**  
Base merchant reliability: fulfillment, dispute/refund health, catalog operational quality, store age, and human-order volume.

**Agentic readiness**  
Readiness to serve agents: setup quality, protocol security (Ed25519 signature), compliance health, and integration freshness.

**Agentic evidence**  
Evidence derived from real agentic transactions (checkout, fulfillment and dispute health per agent, receipt integrity).

**External assurance** bonus · max +10   
Additive bonus (max +10) for external certifications that clear the v4.1 gates. Not a weighted dimension; it is added after the base score.

### Confidence tiers: caps and per-dimension weights

A store's confidence tier is set by its order history, not chosen by the merchant. The tier both caps the maximum score and sets how much each dimension counts.

`—`**insufficient\_data** No applicable components (or a synthetic\_blocked signal): the v4.1 engine emits no value. That does NOT mean the response comes without a score: what happens then is the declared fallback to the v2 engine, and the response says so in its \`engines\` field (\`overall: "v2"\`). Treat it as a DEGRADED result: v2 applies neither the tier caps nor the agentic-evidence checks of v4.1, so it is not comparable with a v4.1 score.

`65`**cold\_start** humanOrders < 5 and agentOrders < 5. With continuous tier interpolation active in production, this cap is not flat: it ranges from 55 (no orders yet) up to 85 as orders accumulate, not a flat 65. Weights: merchantReliabilityBase 0 · agenticReadiness 1 · agenticEvidence 0.

`85`**human\_established** humanOrders ≥ 5 and agentOrders < 5. Weights: merchantReliabilityBase 0.7 · agenticReadiness 0.3 · agenticEvidence 0.

`92`**agent\_verified\_thin** agentOrders between 5 and 19 (interpolation only within the tier). Weights: merchantReliabilityBase 0.45 · agenticReadiness 0.25 · agenticEvidence 0.3.

`100`**agent\_verified** agentOrders ≥ 20 and GMV ≥ 1,000,000 cents (€10k) in 90d and ≥ 20 unique buyers. Without the GMV floor or unique-buyer count → degrades to agent\_verified\_thin. Weights: merchantReliabilityBase 0.35 · agenticReadiness 0.25 · agenticEvidence 0.4.

### Data-quality vocabulary

Every signal behind the score is tagged with one of these six levels, published alongside the number so a reader can tell measured fact from merchant claim.

**`verified`**   
Value derived from cryptographically verified evidence (e.g. TrustReceipt validation outcomes, or certifications that pass the gates).

**`measured`**   
Value from a real, direct measurement.

**`proxy`**   
Value from a related but indirect signal, not a direct measurement of the target (e.g. a store-wide rate used as an approximation of a per-agent metric).

**`self_declared`**   
Value declared by the merchant, not independently verified.

**`not_applicable`**   
No data available for the signal; its weight is redistributed instead of scoring a synthetic value.

**`synthetic_blocked`**   
Defensive fail-closed marker: if any signal were synthetic, the score collapses to insufficient\_data instead of scoring an invented value.

[Read the full trust score story →](https://trusteed.xyz/trust-score)

Legacy v2 · compatibility 

## Legacy v2: 12 weighted components

These weights describe v2 only (0–1 scale), including its fallback path. They do not explain an overall score produced by v4.1. The methodology endpoint retains the fields for compatibility and identifies their scope under legacy\_v2.

**Catalog completeness** Verified   
Ratio of products with all required fields (title, price, images, GTIN) **0.11**

**Catalog freshness** Verified   
Age of last successful catalog sync, normalized 0–1 **0.11**

**Price accuracy** Verified   
Match rate between listed price and live price at checkout **0.12**

**Availability accuracy** Verified   
Stock-status accuracy vs. actual inventory **0.08**

**Policy coverage** Declared   
Fraction of required policy fields filled (return, shipping, privacy) **0.08**

**Checkout success rate** Verified   
Ratio of checkout\_complete / checkout\_start agent events **0.11**

**Fulfillment rate** Verified   
Orders fulfilled on-time over trailing 90 days **0.08**

**Dispute rate** Verified   
Dispute tickets / total orders (lower = better) **0.07**

**Agent satisfaction rate** Verified   
Agent checkout conversion rate (CHECKOUT\_COMPLETE / CHECKOUT\_START events) **0.08**

**Response latency** Verified   
API p95 response latency score (1.0 = < 200ms, 0.0 = > 2000ms) **0.05**

**Review sentiment** Declared   
Average product rating normalized 0–5 → 0–1. Asserted by the merchant, not verified by Trusteed; defaults to 0.5 with no ratings on file **0.05**

**Data consistency** Verified   
Cross-field consistency (price/currency, stock/quantity, GTIN uniqueness) **0.06**

Legacy v2 formula: weighted\_sum\_clamped\_0\_1 · scheduled every 6 hours

Legacy v2 reference 

## Published display-label bands (0–1)

These are the display-label thresholds published by the trust-score endpoint, not permissions to act — a score label, identity verification level and operational status are different fields. Consult the policy of the specific endpoint.

`0.9`**EXCELLENT** Display band only; grants or removes no permission — a separate operational\_status field (active/deprioritized/hidden/suspended) governs search visibility and auto-suspension on its own threshold ladder.

`0.8`**GOOD** Display band only; grants or removes no permission — a separate operational\_status field (active/deprioritized/hidden/suspended) governs search visibility and auto-suspension on its own threshold ladder.

`0.7`**ACCEPTABLE** Display band only; grants or removes no permission — a separate operational\_status field (active/deprioritized/hidden/suspended) governs search visibility and auto-suspension on its own threshold ladder.

`0.5`**CAUTION** Display band only; grants or removes no permission — a separate operational\_status field (active/deprioritized/hidden/suspended) governs search visibility and auto-suspension on its own threshold ladder.

`0`**RESTRICTED** Display band only; grants or removes no permission — a separate operational\_status field (active/deprioritized/hidden/suspended) governs search visibility and auto-suspension on its own threshold ladder.

Evidence and limits 

## What verification establishes

- A JWS signature authenticates the issuer’s record and detects changes to its signed content; it does not verify every source assertion.
- The v2 job runs every 6 hours; the public v4.1 overlay can run on request. No qualified QTSP timestamp is claimed: signed\_at is asserted by the issuer.
- An append-only application table is not, by itself, proof of universal immutability or an independently witnessed history.

If you disagree with your score 

## Dispute process

Merchants can formally appeal a score they believe is wrong.

Endpoint

`POST /api/v1/trust/appeal`

Response SLA

72h

Documentation

[/trust](https://trusteed.xyz/trust)

Questions 

## Frequently asked

### Can a merchant improve their score by self-reporting good numbers?

v4.1 labels signal quality and redistributes weight when evidence is not applicable. In the legacy v2 table, 10 of 12 components are labelled verified (weight 0.87); policy\_coverage and review\_sentiment are declared. These v2 labels do not prove that every input was independently verified.

### How often does the score change?

The legacy v2 component job is scheduled every 6 hours. The public v4.1 overlay can be computed on request. Read the engine, calculated\_at and freshness fields together; in this deployment (coherent engine on) a response is never mixed — overall and breakdown share the same timing.

### Where do I get today's actual number for a store?

Read the signed public response from the merchant’s declared API host: /api/v1/stores/{slug}/trust-score. Check engines, scales and confidence; a signature authenticates the record, not the reliability of every underlying input.

### Is v4.1 (0–100 scale, tiers) live?

Trust Score v4.1 combines three weighted dimensions with an optional external-assurance bonus and confidence-tier caps. Each response's \`engines\` field names the engine actually used: v4.1 for both overall and breakdown, or v2 for both. Two different situations fall to v2 — insufficient\_data (no applicable evidence for that store, a normal business outcome) and v2\_degraded (the overlay is off or failed technically) — read \`confidence\_level\` to tell them apart. A legacy v2 component breakdown does not explain an overall score produced by v4.1.
