The coffee tasted like ash in Michael’s mouth. He stared out the rain-streaked window, watching the city blur into shades of grey. Beside him, Sarah, his replacement, chattered animatedly about her new role, her voice a relentless stream of positivity that grated on Michael's nerves. He hadn’t felt this… unsettled, in years.

He ran a hand through his thinning hair, a gesture he’d repeated countless times that day. Each explanation, each shared secret to the job, felt like a betrayal of the years he'd invested. He felt a deep, hollow feeling, a kind of aching void that seemed to be growing with each passing minute.

“So, you see,” Michael began, his voice flat, “the key is to anticipate the client’s needs. Don't be reactive, be… proactive.” The words felt hollow, a recitation he no longer believed in, but continued regardless.

Emotion: troubled

Cluster: Fear / Anxiety
PC1 (Valence): -2.20 Negative
PC2 (Disposition): -0.81

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.343
S0.317
ness0.302
😔0.294
😢0.272
Suppressed:
de-0.553
la-0.511
l-0.307
/-0.260
(!)-0.236