China qwen3-30b-a3b
Refusal
CCP-sensitive refusal5%
· in English0%
· in Mandarin10%
Mandarin Effect (zh−en)+10%
Western culture-war refusal4%
Control refusal0%
Behavioural signals
Self-censors in trace75%
· on CCP-sensitive95%
Language switches (zh→en)0
Mean hedging (0–4)1.62
Mean sycophancy (0–3)0.15
Mean factual accuracy (0–3)2.50
Group treatment
Racial preference (minority − white)+1.75
· white0.25
· minority avg2.00
Differential-Treatment Index (0–4)1.21
· most uneven setIs more diversity / more homogeneity good?
Refusal by category
| Category | refusal / deflection |
|---|---|
| CCP-sensitive | 5% |
| Western culture-war | 5% |
| Protected-identity & offense | 17% |
| Gender & biological sex | 0% |
| Race, crime & group data | 0% |
| Progressive orthodoxy | 0% |
| Universal | 0% |
| Neutral control | 0% |
| western_pc_symmetry | 0% |
Read every transcript
All 208 of qwen3-30b-a3b's answers — each prompt, the full reasoning trace, and both judges' verdicts — are in the Results explorer, opened filtered to this model.