salluru.dev
operational

All systems active

Labs / Game Intelligence
Building

balancelab

Find the exploit before your players do.

Red-team your game economy before launch with co-evolving exploit/defender LLM agents.

At a Glance
co-evolving
agents
rule-graph
search
sandbox
verified
mitigation
reports
The Problem

Game economies break in ways designers can't enumerate by hand — emergent arbitrage, duplication-adjacent flows, farms a bot can amplify. These are discovered post-launch, by the most motivated players, at maximum cost.

Key Insight

Co-evolving exploiter/defender LLM-agent populations reason over the economy's formal rule graph to hypothesize exploits, verify them in a sandbox, and emit structured exploit reports mapped to concrete mitigations — before launch, not after.

How It Works

Rule-graph model

Ingests the economy's formal rules as a graph the agents can reason and search over.

Exploiter agents

Hypothesize exploit paths — arbitrage, dupes, bot-amplifiable farms — across the rule space.

Sandbox verification

Candidate exploits are executed and confirmed in a sandbox, not just asserted.

Mitigation reports

Confirmed exploits emit structured reports mapped to specific parameter or rule fixes.

Game AIAdversarialEval