Available models and recommended scores
At this time, we have deployed a few different models to help verify unique humans. We plan to expand this list of models greatly, first by exploring additional Sybil detection models for several L2 partners, and then looking into other web3 verticals and reputation signals.
We currently offer the following models via the Model Based Detection API:
- Aggregate unique humanity score -
aggregate - Ethereum (L1) unique humanity model -
ethereum_activity - NFT (L1) unique humanity model -
nft - Arbitrum unique humanity model -
arbitrum - Optimism unique humanity model -
optimism - Polygon unique humanity model -
polygon - zkSync unique humanity model -
zksync
Each model will assign a score of 0 - 100 to wallet addresses that are checked against it. A score of 0 represents likely Sybil, while a score of 100 represents likely human.
For each model, we will provide a table that describes the different score thresholds you can use to gate access or classify addresses. We define the columns of those tables here:
- % of qualifying verified humans - This metric represents the number of Passport users who qualified for the score threshold and scored a 20 or higher with the Stamp-based verification system.
- % of verified Sybil penetration - This metric represents the number of verified Sybils from our list that were able to qualify for the score threshold.
Please note that the percentages included in these tables do not necessarily translate to the percentage of participation for each individual partner.
We calculate the percentage of qualifying humans and Sybils at various thresholds, derived from a curated list of collected addresses labeled as either Sybils or humans.
If you would like more details on the score or these thresholds, please fill out the following form:
Model Based Detection interest form (opens in a new tab)
Aggregate unique humanity model
The aggregate unique humanity model score is the current default for the Model Based Detection API, as it is broadly relevant to a variety of different ecosystems operating within the EVM ecosystem.
This model integrates individual chain model scores (Ethereum, zkSync, Optimism, Arbitrum, and Polygon) to generate the final aggregate score, assigning a score between 0 - 100, with 0 being likely Sybil or not having enough transaction history, and 100 being confidently human.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 97% | 14% |
| 25+ | 96% | 12% |
| 50+ | 94% | 9% |
| 75+ | 89% | 5% |
ETH (L1) unique humanity model
This model looks at the specified wallet addresses' ETH mainnet transaction history, compares it against 50+ different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 97% | 14% |
| 25+ | 96% | 12% |
| 50+ | 94% | 9% |
| 75+ | 89% | 5% |
NFT (Ethereum L1) unique humanity model
Similarly to the ETH (L1) unique humanity model, this model looks at the specified wallet addresses' NFT transaction history on Ethereum mainnet, compares it against different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 98% | 11% |
| 25+ | 97% | 9% |
| 50+ | 95% | 6% |
| 75+ | 92% | 4% |
Arbitrum unique humanity model
This model looks at the specified wallet addresses' Arbitrum transaction history, compares it against different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 99% | 21% |
| 25+ | 98% | 14% |
| 50+ | 94% | 7% |
| 75+ | 92% | 6% |
Optimism unique humanity model
This model looks at the specified wallet addresses' Optimism transaction history, compares it against different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 98% | 14% |
| 25+ | 97% | 10% |
| 50+ | 96% | 5% |
| 75+ | 94% | 4% |
Polygon unique humanity model
This model looks at the specified wallet addresses' Polygon transaction history, compares it against different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 98% | 15% |
| 25+ | 97% | 11% |
| 50+ | 94% | 7% |
| 75+ | 92% | 6% |
zkSync unique humanity model
This model looks at the specified wallet addresses' zkSync Era transaction history, compares it against different features, and assigns it a score of 0 - 100.
Refer to the following table to identify which score threshold you might use to protect access or classify wallet addresses in your ecosystem:
| Score threshold | % of qualifying verified humans | % of verified Sybil penetration |
|---|---|---|
| 15+ | 95% | 22% |
| 25+ | 94% | 16% |
| 50+ | 92% | 8% |
| 75+ | 88% | 7% |
Models coming soon
The Passport team is also working on adding an aggregate model to this endpoint, and is considering several other models that look at unique humanity and reputation signals for different ecosystems.
To request a new model, please fill out this form: Model Based Detection feedback form (opens in a new tab)
Next Steps
Learn how to use these models and score thresholds by working through our tutorial, or reviewing our API reference.