The authors introduce Stellar Colosseum, a model-agnostic harness designed to allocate inference across long-horizon research problems in mathematics and theoretical computer science. The system explores alternative strategies before proof construction, uses a readiness gate to determine when to decompose routes, and represents proof plans as interdependent subproblems.
Colosseum generates candidates in parallel, attacks them with targeted falsification, and combines results through overlapping random-sample tree aggregation. This workflow is integrated into Google Antigravity's Teamwork framework as the Long Proof pattern. Evaluations on TCS-Bench show 71.0% accuracy using Gemini 3.1 Pro and Gemini 3.7 Flash. Additionally, a proof-oriented pipeline with execution feedback solved 218 of 222 Codeforces problems.
The harness addresses the unreliability of language models on complex research tasks by managing sequences of uncertain decisions through structured verification and aggregation.