Performance Benchmarks

ABSCORE™ is engineered for Zero-Latency Governance. By utilizing a Rust-based Kernel compiled to WebAssembly (WASM), we achieve sub-millisecond policy evaluation, making it suitable for high-frequency agentic systems.

Key Metrics

Component	Metric	Target	Result (p99)
Kernel WASM	Policy Evaluation	< 5ms	1.2ms
Magic Proxy	Request Overhead	< 20ms	8.4ms
Total Pipeline	End-to-End Latency	< 50ms	32.0ms

[!NOTE] Tests conducted using a sustained load of 100 requests/sec on Cloudflare Edge nodes.

High-Resolution Trace

Every decision logged by ABSCORE™ includes high-resolution telemetry. The following trace shows a typical authorization cycle:

sequenceDiagram
    participant Agent
    participant Proxy as Magic Proxy (TS)
    participant Kernel as ABS Kernel (WASM/Rust)

    Agent->>Proxy: Tool Call Request
    Proxy->>Proxy: JSON Parsing (0.4ms)
    Proxy->>Kernel: evaluate(payload)
    Kernel->>Kernel: Regex Engine Match (0.8ms)
    Kernel->>Kernel: Risk Scoring (0.2ms)
    Kernel->>Kernel: Signed Proof Generation (0.2ms)
    Kernel-->>Proxy: Decision Object
    Proxy-->>Agent: Allowed/Denied + Proof Hash

Testing Protocol

We use Artillery for sustained load testing and Vitest for unit-level latency tracking within the WASM sandbox.

1. Kernel Stress Test

We evaluate 50+ complex regex policies against 1KB payloads.

Concurrent Users: 500
Duration: 60 seconds
Success Rate: 100.0%

2. Edge Overhead

The Magic Proxy adds minimal overhead to standard LLM calls. In most cases, the governance latency is smaller than the network jitter of the LLM provider (OpenAI/Gemini).

ABS CORE Pro Tip: Use X-ABS-Shadow-Mode: true to measure performance impact in your environment before enabling strict blocking.

Performance & Benchmarks