ABS Core v3.5.0
Technical

Performance & Benchmarks

Benchmark methodology and measurement boundaries for ABS Core enforcement latency.

Performance & Network Reality

ABS Core is engineered for high-frequency environments, but safety always respects the laws of physics.

[!WARNING] Industrial Disclaimer: The benchmarks below reflect deployment on Cloudflare Workers (Edge Computing) nodes. For Self-Hosted Sovereign Enforcement (WASM Sidecar), benchmarks vary based on host hardware, typically targeting < 1.7ms processing overhead. Self-hosted benchmarks are being audited for Q2 Release.

Verified Benchmarks (Edge Runtime)

MetricProcessing (WASM)Total Overhead (Sidecar)Total Overhead (Proxy Mode)
Logic Evaluation< 0.8ms< 1.5ms~5ms - 20ms (Network dependent)
Intent Parsing< 0.2ms< 0.5ms+Network RTT
Max Throughput15,000+ TPS8,500+ TPS2,000+ TPS

The Sovereignty/Performance Trade-off

  1. Proxy Mode (Gateway): Easiest to integrate (REST/gRPC), but subject to network latency and potential bypass if the host is compromised.
  2. Sovereign Sidecar (Local): The Standard-Gold for security. By running ABS as a sidecar process on the same host, network latency is eliminated, and governance becomes a non-optional system gate.

[!IMPORTANT] For Banking, Healthcare, and Defense applications, Sidecar Mode is mandatory to eliminate the "Network Gap" and ensure zero-bypass safety.

[!IMPORTANT] In 98% of cases, the ABS Core enforcement latency is lower than the standard deviations of the upstream LLM network latency. Governance is practically invisible.

Nesta página