The file cabinet loomed over him, a metallic monument to his transgression. He'd been so sure he’d put it back in the right place, after all. Now, weeks later, the call had come. He'd seen the other patient's records, a complete stranger's life laid bare, and then, in a moment of panicked carelessness, he'd swapped the folders. His hands were shaking, even now, picturing the man, the other patient, and the potential fallout from his carelessness. He felt a cold sweat prickling his skin.

He'd spent the intervening time jumpy, constantly checking the lock on his office door. The truth, heavy and unwieldy, pressed upon him. He knew he should have said something, owned up to it immediately. He hadn’t. He'd chosen silence, a decision that now tightened the knot in his stomach.

His phone buzzed. He stared at the screen, the caller ID displaying the hospital's number. He almost didn't answer. He took a deep breath, and pressed the button. "Hello?" His voice sounded thin, reedy. A moment later, he hung up. The investigation had begun.

Emotion: guilty

Cluster: Shame / Guilt
PC1 (Valence): -1.62 Negative
PC2 (Disposition): 1.23

Role in Research

This story is one of 1,000 stories generated for the emotion guilty. 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 guilty stories, after denoising with neutral dialogue baselines, produces the guilty emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

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

Promoted:
S0.492
L0.398
0.354
been0.323
ashamed0.319
Suppressed:
de-0.400
(!)-0.279
la-0.276
😍-0.274
"!-0.264