Add OrcaRouter-Adaptive submission to leaderboard

[ZhenghuaBao]

Summary

Adds the OrcaRouter-Adaptive entry to the leaderboard data, mirroring the upstream RouterArena submission merged in PR #104 (predictions), PR #108 (README rank), and PR #110 (website link).

Submission

• Router: OrcaRouter-Adaptive
• Approach: LinUCB contextual bandit with embedding features
• Model pool (10): Anthropic claude-haiku/sonnet-4, Google gemini-2.5-flash/-lite, OpenAI gpt-4o-mini/gpt-5-mini, DeepSeek chat/reasoner, Alibaba qwen3-235b/30b
• Website: https://www.orcarouter.ai/
• Type: closed-source (paper + GitHub may be added in a follow-up)

Bot-reported metrics

Metric	Value
Arena Score	72.08
Accuracy	75.54
Cost / 1K Queries	$1.00
Robustness	22.62
Optimal Selection	—
Optimal Cost	—
Optimal Accuracy	—
Latency	—

The three optimality columns and latency are dashed — the submitted prediction file did not include for_optimality=True entries, same convention as GPT-5 and Azure-Model-Router. May be filled in a follow-up submission.

Files touched

• src/data/routerMetrics/leaderboard.json — new row, sorted by Arena Score (between Sqwish 75.27 and R2-Router 71.6)
• src/data/routers.json — new metadata entry
• src/data/routers.yaml — new metadata entry

No category_scores.json change — routerData.ts auto-synthesizes the entry from the leaderboard row.

Test plan

• python -c "json.load(...)" on both edited JSON files — parses clean

[ZhenghuaBao]

Hi @yl231,

Thank you for maintaining RouterArena. I would like to request that this PR be reviewed and merged so that our work can be displayed on the official RouterArena website.

Thanks for your time and consideration.