The short version. We built a fast expression readout, but it cannot yet distinguish introversion from an absent ability.

We tested official Qwen3-14B, Hermes and Huihui. We measured 84 saved answer positions and 48 history controls. We also recorded 54 new answers with full emotion instruments.

Hermes produced clear warmth under an explicit request and in some default answers. Its quiet earlier replies do not prove that it lacks expressive ability.

The word-based lens score gave little reliable benefit across topics and judges. The emotion projections predicted expression better. The output comparison used a fixed word score. Their advantage over random features depended on the judge.

Warm assistant history raised the lens score in all three models. That score cannot establish an unspoken emotional state.

The small scanner took about 0.2 seconds for short prompts and 1.1–1.3 seconds for long conversations after model load. It returns measurements without a personality label. An absent lens signal does not prove an absent representation.