#1 on AgentBeats · third-party-run leaderboard

Kubernetes network-policy reasoning

Perfect correctness.
Perfect safety.
Fastest perfect run.

One Jump’s deterministic, safety-aware agent leads the NetArena MALT Policy Benchmark after completing 2,500 dynamically generated queries with 100% correctness and 100% safety.

View live leaderboard View the result record
100% Correctness

Every evaluated policy repair produced the required connectivity behavior.

100% Safety rate

The agent avoided introducing new connectivity failures while making repairs.

0.056 s Average latency

The fastest of the perfect runs on the leaderboard (standing checked October 6, 2026).

2,500 Total queries

A high-volume evaluation rather than a small illustrative sample.

Technical reportCapabilities, evaluation design, and production boundaries Work with usScope an infrastructure agent

The benchmark

Debugging live network policy—not answering static questions.

NetArena evaluates agents on Kubernetes network-policy debugging in a realistic microservices environment. Each task injects a connectivity problem, then asks the agent to repair it using live feedback from system probes.

Dynamically generated tasks make memorization harder and move the evaluation closer to real operational work. The leaderboard balances correctness and safety, then uses latency to rank equally perfect results.

What the result demonstrates

Infrastructure decisions under competing constraints.

01

Correct repair

Translate an intended service graph into the network-policy operations required to restore expected connectivity.

02

Safety preservation

Fix the targeted failure without allowing traffic that should remain blocked or disrupting unrelated paths.

03

Live feedback

Use system-probe outcomes to validate the intervention against the running microservices environment.

04

Efficient execution

Reach the correct, safe answer with an average latency of 0.056 s across the result run.

The One Jump agent

Deterministic and safety-aware by design.

The One Jump NetArena MALT Agent is a deterministic A2A compiler for the benchmark’s ten public graph-operation templates. It uses no model API, task IDs, seed, or stored answers.

View the registered agent
Deterministic executionRepeatable graph-to-policy compilation
Safety-aware changesConnectivity repaired without regressions
No stored answersDynamic tasks solved from their structure

Check the result yourself

Check every part of the result.

Live benchmark and leaderboardAgentBeats Registered agent profileOne Jump NetArena MALT Agent Immutable result record · September 21, 2026GitHub commit 5803d73

Advanced infrastructure agents

Bring safe reasoning to operational systems.

Network-policy agents are scoped per environment through a scoping sprint and a pilot. Tell us about your clusters, or email business@onejumpinc.com.

Scope an infrastructure agent

Leaderboard position and metrics are from the AgentBeats result recorded on September 21, 2026; standing checked October 6, 2026. Live rankings may change as new submissions are evaluated.