Table 6.

Expected LLM-backbone sensitivity of R-GCRED on DAIC-WOZ.

R-GEE backbone	Developer	Model type	Macro-F1 ↑	AUC ↑	Sensitivity ↑	ECE ↓
Llama 3.1 8B Instruct	Meta	Open-weight	0.81	0.86	0.79	0.052
Gemini 2.5 Flash	Google	Proprietary	0.83	0.88	0.81	0.043
Claude Sonnet 4.6	Anthropic	Proprietary	0.83	0.88	0.81	0.041
GPT-4.1	OpenAI	Proprietary	0.84	0.89	0.82	0.038

↑ indicates higher-is-better and ↓ indicates lower-is-better.