SetRaceAgent TrueSkill μ=27.3±3.3

fourth-highest TrueSkill rating

Source paper

extracted_from

(2026) · Robert Müller · Clemens Müller

claim

Two heuristic code agents (TrackerAgent and SetRaceAgent) outperform most tested LLMs.
supports
Calibration that conditional logic can beat cost-efficient LLMs in this setting.

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

EconomyAgent TrueSkill μ=22.1±3.5finding0.844
EconomyAgent TrueSkill rating
DS-v3.2 TrueSkill μ=23.9±2.7finding0.836
DeepSeek v3.2 TrueSkill rating
TrackerAgent TrueSkill μ=28.7±3.6 in 98 canonical gamesfinding0.833
second-highest TrueSkill rating
TrackerAgent TrueSkill μ=28.7 ± 3.6, 53.6% win ratefinding0.821
Best code agent outperforming six of seven LLMs.
GPT5.4-N TrueSkill μ=22.6±2.7finding0.810
GPT-5.4 Nano TrueSkill rating
G2.5-FL TrueSkill μ=20.5±2.9finding0.807
Gemini 2.5 Flash Lite, lowest TrueSkill rating
Haiku TrueSkill μ=21.8±2.8finding0.794
Claude Haiku 4.5 TrueSkill rating
G3.1-FL TrueSkill μ=28.0 ± 2.9, 44.9% win ratefinding0.786
Second-best LLM, competitive with TrackerAgent.