The email shimmered on her phone screen, the subject line a blunt instrument: "Student Spotlight: Application Essay Example." Sarah froze, her fingers going numb. She scrolled through the university's website, each word of her essay mocking her from the page. It was *her* words, all right, the painstakingly crafted prose she’d slaved over. A cold sweat prickled her skin.

Her gaze darted around the coffee shop, as if expecting someone to recognize her, to point a finger. The essay, a tale of overcoming adversity, had been a lie. Or, at least, a highly embellished version of the truth. She’d stretched the details, borrowed a phrase or two from a friend's experience. Now, it was out there, a monument to her… fabrication. She hadn’t slept well since, a knot twisting in her stomach each time she remembered what she'd done.

The barista called her name. Sarah jumped, fumbling for her wallet. "Thanks," she mumbled, the word barely audible. She bolted for the exit, the essay's perfect sentences echoing in her skull.

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