🤖 3D Agent Plan-Act-Critique Retry Loop: Absorbing Markov Chain, Attempt Budget, Silent Failures and Token Cost
An agent line built in Blender: crates are tasks that pass Plan, Act (robot arm) and Critique (lens) stations, rejected ones ride the return belt; closed-form outcome probabilities, expected attempts, latency and token cost from an absorbing Markov chain, checked by a linear solve and a 200 000-task Monte Carlo.
⚙ Under the hood
An agent line built in Blender: crates are tasks that pass Plan, Act (robot arm) and Critique (lens) stations, rejected ones ride the return belt; closed-form outcome probabilities, expected attempts, latency and token cost from an absorbing Markov chain, checked by a linear solve and a 200 000-task Monte Carlo.
LLM AgentsPlan-Act-CritiqueRetry BudgetAbsorbing Markov ChainCritic Error RatesLatency and CostAgent ReliabilityMonte Carlo
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install