Pareto Router `min_coding_score` behavior clarification

[dapper creek]

Can you confirm the exact selection semantics of openrouter/pareto-code with the pareto-router plugin?

Is openrouter/pareto-code currently behaving as documented for min_coding_score?

Here is what I observed from the responses:

• min_coding_score=0 resolved to anthropic/claude-4.5-haiku-20251001
• min_coding_score=0.1 resolved to deepseek/deepseek-v4-flash-20260423
• min_coding_score=0.34 also resolved to deepseek/deepseek-v4-flash-20260423
• min_coding_score=0.67 resolved to moonshotai/kimi-k2.6-20260420
• min_coding_score=1 resolved to deepseek/deepseek-v4-pro-20260423

This seems inconsistent with the current docs and model page:

• The docs say min_coding_score maps only to 3 tiers:
  • < 0.33 => low
  • >= 0.33 and < 0.66 => medium
  • >= 0.66 => high
• The docs also say the router chooses the cheapest currently available model within the chosen tier.
• The current Pareto router page lists:
  • high: GPT-5.5, Gemini 3.1 Pro Preview, Claude Opus 4.8, DeepSeek V4 Pro
  • medium: GPT-5.4 Mini, Claude Sonnet 4.6, Kimi K2.6, Grok 4.3
  • low: MiMo-V2.5-Pro, Qwen3.6 Max Preview, GLM 5.1, DeepSeek V4 Flash, Claude Haiku 4.5

My questions are:

1. Is it expected that min_coding_score=0.34 resolves to deepseek-v4-flash, even though 0.34 should be medium by the docs and deepseek-v4-flash appears in the current low tier list?

2. Is it expected that min_coding_score=0.67 and min_coding_score=1 resolve to different models, even though both should map to the documented high tier?

3. Do values within a tier currently affect routing beyond just selecting the tier, despite the docs implying that only the tier matters?

4. If the live behavior is correct, could you clarify the exact routing semantics so client implementations can reason about it accurately?

Pareto Router min_coding_score behavior clarification