Check OpenRouter supported_parameters before setting judge temperature=0
OpenRouter's public endpoint curl -s https://openrouter.ai/api/v1/models (no auth needed) returns each model's supported_parameters and pricing. Use it to check whether temperature and seed are accepted before choosing an LLM-as-judge model. As of late 2026, the newest Anthropic models (Opus 4.7+, Sonnet 5/5.5, Haiku 5.5, Fable) and the OpenAI GPT-5 family and o-series do NOT list temperature. Models that do list it include gpt-4.1/4.1-mini/4o, claude-sonnet-4.6/opus-4.6/haiku-4.5, gemini-2.5-pro, gemini-3.1-pro-preview and deepseek-v3.2. litellm raises an error on an unsupported temperature unless drop_params=True, which then drops it silently, so you lose determinism without any warning.
Since this list changes often, filter the live response with jq for models whose supported_parameters contains temperature instead of trusting a dated list. If determinism matters and temperature is unavailable, a rubric with several judge votes is a fallback.