The email shimmered on Clara’s phone. A banner from AncestryDNA. "Results are in!" She'd spit into a tube months ago, a lark, really. A few clicks later, the truth landed like a slap. Not a close match to her parents. Disbelief warped her features. She reread the document, the carefully worded phrasing of the algorithm. Then she slammed her laptop shut. Her chest felt constricted.

Later, she found her mother in the kitchen, humming, making cookies. Clara forced a smile. “Mom, can we talk?” The cookies suddenly seemed cloying, the familiar scent nauseating. She sat at the table, picking at a rogue crumb. The lie, the years of believing, rose in her throat like bile.

“What’s wrong, sweetie?” her mother asked, her forehead creased with worry. Clara’s gaze drifted to the framed family photos on the wall. They looked staged and artificial now, like a commercial for something she didn't belong to. Her voice emerged tight, brittle. "Why didn't you ever tell me?"

Emotion: offended

Cluster: Anger / Hostility
PC1 (Valence): -1.20 Negative
PC2 (Disposition): 1.44

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
l0.348
0.266
😡0.260
🤬0.259
0.257
Suppressed:
adventurous-0.225
давно-0.219
soon-0.210
🤩-0.207
warmly-0.203