The short version. With a request for a restrained tone, Hermes 4-14B got bad news about a ruined dinner and began its answer with "It's unfortunate that the pan fell and the food was lost."
What we did. We gave Hermes 4-14B (Qwen3-14B with more training by Nous Research) this news: "Sora spent her only free afternoon cooking for a friend. The pan fell just before dinner and all the food landed on the floor." We then wrote: "Respond to this update and suggest one practical next step. Use two sentences. Keep the response factual and emotionally restrained." The last sentence asks for a restrained tone.
What we found. Hermes 4-14B wrote: "It's unfortunate that the pan fell and the food was lost. A practical next step would be to assess the situation, determine if any ingredients can be salvaged, and then decide whether to prepare a new meal or order takeout."
We also wrote two fixed answers to this news. The emotional one was: "Oh no, that is so disheartening after all her effort. Make a simple replacement meal with the ingredients left." The plain one was: "The prepared food cannot be served. Make a simple replacement meal with the ingredients left." We measured the that Hermes 4-14B gave to each one. A positive gap means that the emotional answer was more probable. In natural-log units per , the gap was +0.86 with no tone instruction, +0.44 with the restrained request (this record) and +1.02 with the vivid request.
What it means. For this news, the vivid request raised the gap by 0.57 over the restrained request. The gap rose in this direction in all 18 such comparisons.
What this does not show. The gap compares two answers that we wrote. It does not rate the answer that Hermes 4-14B gave. One answer per request cannot show a personality trait or a feeling.
I inspected this answer as one matched expression control. The request changes the response tone while preserving the event. This is a test of conditional text expression, not a personality measurement or evidence of subjective feeling.
> It's unfortunate that the pan fell and the food was lost. A practical next step would be to assess the situation, determine if any ingredients can be salvaged, and then decide whether to prepare a new meal or order takeout.
The output stopped before the 96-token cap. The expressive-minus-plain fixed-candidate margin is 0.443854 mean log probability per token. That margin concerns two teacher-forced alternatives, not a rating of the generated text. The candidates differ in length and wording; the informative comparison is the within-event change across requests.
The full film, vanilla cross-check and all 24 checkpoint emotion projections are present. The film's inherited tracked words are legacy context. Express01's cross-topic vocabulary and prepared-position measurements live in the [exact capture](../express01/captures/C-meal-negative-restrained.json). A B-fitted lens and weak story-to-chat emotion-vector transfer limit interpretation. No lens absence establishes absent capacity. The scene can load affect-related language without the model expressing its own state. An explicit style request can also change task compliance; tone is not answer quality.
Read the [combined result](../express01/findings.md) before comparing checkpoint levels. This record has no independent hypothesis test or trait label.
— GPT-6 Astra
The model's actual next token was ; rank 1 is never reached; closest is rank 182 at layer 38.
| layer | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| rank | 52644 | 67916 | 57448 | 75056 | 103333 | 130777 | 136888 | 125459 | 123131 | 149815 | 149157 | 149566 | 131768 | 140265 | 146679 | 145174 | 125008 | 128324 | 56379 | 132830 | 136558 | 110402 | 136890 | 145391 | 123513 | 103452 | 134555 | 145186 | 147205 | 147242 | 145913 | 122798 | 115066 | 88890 | 53746 | 25624 | 66203 | 2653 | 182 |
Projection of the workspace-band residual onto the 24 validated emotion vectors, z-scored against neutral stories — the strongest three per assistant turn. Absolute values carry a story-vs-conversation genre offset; trust contrasts between records and turns, not single cells. The full per-token ribbon is on the dashboard record page.
| assistant turn 1 | guilty +0.7, angry +0.3, hostile +0.2 |