OpenRank SDK

A command-line SDK for interacting with the OpenRank AVS (Actively Validated Service) - a decentralized trust and reputation system that runs PageRank-style algorithms in Trusted Execution Environments (TEEs).

What is OpenRank?

OpenRank is a decentralized system for computing trust scores and reputation rankings using the EigenTrust algorithm (a variant of PageRank). It operates within Trusted Execution Environments to ensure computational integrity and provides censorship-resistant, verifiable trust computations.

The system works with:

• Trust data: Relationships between entities with trust weights
• Seed data: Initial trust scores for bootstrapping the algorithm
• Score computation: Running EigenTrust to generate final trust scores

Installation

Build the SDK from source:

cd sdk
cargo build --release

The binary will be available at target/release/openrank.

Commands

Initialisation

Initialise the workspace with example datasets and .env file:

openrank init ./my-workspace

This command will create a folder with following structure:

• trust/
• seed/
• .env

.env file contains a placeholder for your mnemonic phrase:

MNEMONIC="add your mnemonic phrase here"

Operations

compute-request

Submit a computation request using trust and seed data from local folders.

openrank compute-request <TRUST_FOLDER_PATH> <SEED_FOLDER_PATH>

Arguments:

• TRUST_FOLDER_PATH - Path to folder containing trust CSV files
• SEED_FOLDER_PATH - Path to folder containing seed CSV files

Example:

openrank compute-request ./trust_data ./seed_data

compute-watch

Monitor and watch for computation results by compute ID.

openrank compute-watch <COMPUTE_ID> [--out-dir <OUT_DIR>]

Arguments:

• COMPUTE_ID - The computation ID to monitor

Options:

• --out-dir <OUT_DIR> - Output directory for results (optional)

Example:

openrank compute-watch abc123 --out-dir ./results

download-scores

Download computed scores for a specific computation.

openrank download-scores <COMPUTE_ID> [--out-dir <OUT_DIR>]

Arguments:

• COMPUTE_ID - The computation ID to download scores for

Options:

• --out-dir <OUT_DIR> - Output directory for downloaded scores (optional)

Example:

openrank download-scores abc123 --out-dir ./scores

Local Operations

compute-local

Run OpenRank computation locally using trust and seed CSV files.

openrank compute-local <TRUST_PATH> <SEED_PATH> [OUTPUT_PATH]

Arguments:

• TRUST_PATH - Path to trust CSV file
• SEED_PATH - Path to seed CSV file
• OUTPUT_PATH - Output path for computed scores (optional)

CSV Format:

• Trust CSV: from_id,to_id,trust_weight
• Seed CSV: peer_id,score

Example:

openrank compute-local trust.csv seed.csv scores.csv

verify-local

Verify computed scores against trust and seed data locally.

openrank verify-local <TRUST_PATH> <SEED_PATH> <SCORES_PATH>

Arguments:

• TRUST_PATH - Path to trust CSV file
• SEED_PATH - Path to seed CSV file
• SCORES_PATH - Path to scores CSV file to verify

Example:

openrank verify-local trust.csv seed.csv computed_scores.csv

Data Formats

Trust CSV Format

from_id,to_id,trust_weight
alice,bob,0.8
bob,charlie,0.6
charlie,alice,0.9

Seed CSV Format

peer_id,score
alice,1.0
bob,0.5
charlie,0.3

Scores CSV Format

peer_id,score
alice,0.45
bob,0.32
charlie,0.23

Or use AWS credential files and profiles as per standard AWS CLI configuration.

Examples

Complete Local Workflow

1. Prepare your data files:

   echo "alice,bob,0.8" > trust.csv
   echo "bob,charlie,0.6" >> trust.csv
   echo "alice,1.0" > seed.csv
   echo "bob,0.5" >> seed.csv
   

2. Run local computation:

   openrank compute-local trust.csv seed.csv scores.csv
   

3. Verify the results:

   openrank verify-local trust.csv seed.csv scores.csv
   

Distributed Computation Workflow

1. Submit computation request:

   openrank compute-request ./trust_folder ./seed_folder
   

2. Monitor computation:

   openrank compute-watch <compute_id> --out-dir ./results
   

3. Download results:

   openrank download-scores <compute_id> --out-dir ./final_scores
   

Algorithm Details

OpenRank implements the EigenTrust algorithm with the following key features:

• Pre-trust weighting: Uses a 0.5 weight for seed trust values
• Convergence threshold: Stops iteration when delta < 0.01
• Reachability filtering: Only includes peers reachable from seed nodes
• Normalization: Ensures trust distributions sum to 1.0

The algorithm iteratively computes trust scores until convergence, providing a robust measure of reputation in decentralized networks.

Getting Help

openrank --help
openrank <command> --help

For more information about the OpenRank system and TEE deployment, see the main project documentation.