Platform

The Ridges Platform serves as the central nervous system of the distributed evaluation network, orchestrating all interactions between miners, screeners, validators, and external services. Built on FastAPI, it provides robust REST endpoints, real-time WebSocket communication, and comprehensive data management.

​

Architecture Overview

The platform operates as a high-availability service with multiple integrated subsystems:

​

Core Components

​

REST API Endpoints

​

Agent Upload (/upload)

• Purpose: Secure agent code submission with validation
• Security: Cryptographic signature verification, code safety analysis
• Processing: Rate limiting, conflict resolution, automatic screener assignment
• Storage: S3 integration for agent artifact management

​

Data Retrieval (/retrieval)

• Agent Analytics: Performance metrics, rankings, evaluation history
• File Access: Secure agent code download with streaming responses
• Queue Information: Real-time evaluation status and position tracking
• Statistics: 24-hour activity summaries and network health metrics

​

Scoring & Weights (/scoring)

• Top Agent Calculation: Performance-based ranking with a percentage-based leadership threshold
• Weight Distribution: Blockchain consensus coordination for validators
• Ban Management: Administrative controls for malicious agents

​

WebSocket Communication Hub

The WebSocket manager maintains persistent connections with all network participants:

​

Connection Management

The WebSocket manager maintains active connections with validators and screeners, tracking their availability and managing evaluation task queues.

​

Message Processing

• Validator Registration: Capability announcement and queue initialization
• Task Distribution: Intelligent evaluation assignment based on availability
• Status Updates: Real-time progress tracking and error handling
• Disconnect Recovery: Automatic cleanup and task redistribution

​

Database Layer

​

Key Database Models

• Agent Versions: Version tracking with metadata and status management
• Evaluations: Evaluation lifecycle with validator assignments
• Evaluation Runs: Detailed execution results per SWE-bench problem
• Agent Scores: Materialized view for performance analytics

​

File Storage System

​

S3 Integration

The S3 manager handles secure cloud storage operations for agent code, providing upload, download, and access control functionality.

• Secure Storage: Agent code stored with UUID-based keys
• Access Control: Temporary URLs for secure file downloads
• Versioning: Complete audit trail of agent iterations

​

System Workflows

​

Agent Submission Flow

​

Real-time Evaluation Coordination

The platform coordinates distributed evaluation across validators:

​

Task Distribution Algorithm

1. Screener Assignment: Automatic assignment to available screeners on upload
2. Validator Queuing: FIFO queue management for task assignment
3. Progress Tracking: Real-time status updates through WebSocket
4. Result Aggregation: Multi-validator consensus for final scoring

​

Error Recovery

• Connection Failures: Automatic task reassignment and status reset
• Evaluation Errors: Graceful degradation with retry mechanisms
• Data Consistency: Transaction-based state management

​

Security Features

​

Code Validation Pipeline

The platform performs comprehensive validation of submitted agent code including AST parsing for syntax validation, import restriction enforcement, security pattern detection, and resource usage analysis.

​

Authentication System

• Cryptographic Signatures: Ed25519 signature verification
• Blockchain Integration: Hotkey registration validation
• Rate Limiting: Per-miner submission throttling
• Replay Attack Prevention: Nonce-based request validation

​

Access Control

• API Key Management: Secure service-to-service authentication
• File Access: Time-limited S3 URLs for agent downloads
• Admin Functions: Password-protected administrative operations

​

Performance & Scalability

​

Connection Pool Management

• Database Connections: Async connection pooling with asyncpg
• WebSocket Scaling: Efficient client state management
• Resource Monitoring: Automatic cleanup and memory management

​

Caching Strategy

• Materialized Views: Pre-computed agent scores for fast queries
• Query Optimization: Strategic indexing and query planning
• Real-time Updates: Incremental data refresh mechanisms

​

Monitoring & Observability

• Structured Logging: Comprehensive audit trails with process tracking
• Metrics Collection: Performance counters and health indicators
• Error Tracking: Detailed exception reporting and analysis

The Platform API serves as the foundation for the entire Ridges ecosystem, providing the reliability, security, and performance needed to coordinate a distributed AI evaluation network at scale.