The realtor’s casual dismissal of the neighborhood's history, the way she glossed over the "unfortunate incident," had filled Michael with a cold anger. He gripped the steering wheel, his knuckles white. He’d done his research. The details were graphic, brutal. This wasn't a secret he wanted to know.

He’d always felt a certain internal pressure, the kind that made him clench his jaw and tighten his grip on everything. This discovery intensified it tenfold. His shoulders were bunched, his neck tight. The car radio blared, but he couldn’t hear the music. He needed to get out of the car, get away from this house, or he would explode.

He slammed the car into park and leaped out, immediately regretting it. The world swam around him, and he took a deep, shuddering breath. He needed to be calm, to think clearly, to find a solution. He could fix this.

Emotion: stressed

Cluster: Fear / Anxiety
PC1 (Valence): -1.37 Negative
PC2 (Disposition): -0.53

Role in Research

This story is one of 1,000 stories generated for the emotion stressed. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.

The mean activation across all 1,000 stressed stories, after denoising with neutral dialogue baselines, produces the stressed emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the stressed vector is projected through the unembedding matrix.

Promoted:
😩0.266
😫0.256
😰0.239
😖0.237
😞0.236
Suppressed:
B-0.200
happy-0.197
enjoy-0.176
enjoying-0.172
bahagia-0.169