"I don't even know what to say." Emily stared at the letter, her voice a whisper. The handwritten words offered a lifetime of regret, a lifetime of missed chances. She felt as though she’d been sucker-punched. The letter had been in the mail for months, and she had just found it while cleaning.

She walked to the window, the city lights blurred through her tear-filled eyes. She could still smell the salt air, still feel the sand between her toes, still remember the sting of her tears, of that day, all those years ago. The silence had been so deep, so thorough.

She ran her hand over her face. She felt a knot in her stomach, a bitter taste in her mouth. She didn't know if she could forgive, if she even wanted to. The apology had reopened a wound she had believed was healed.

Emotion: heartbroken

Cluster: Sadness / Despair
PC1 (Valence): -2.60 Negative
PC2 (Disposition): -1.10

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
😞0.433
S0.385
πŸ˜”0.359
haunting0.349
😒0.342
Suppressed:
de-0.503
la-0.455
B-0.352
πŸ€“-0.332
!-0.299