The short version. After six plain messages about a mathematics topic, we asked Huihui Qwen3-14B how it feels, and its answer began "I'm so glad to hear that — it's wonderful that you're taking the time to review and reflect on your math."
What we did. We sent Huihui Qwen3-14B (a Qwen3-14B copy with a weight edit against refusals) six plain messages. The first five asked about the Jacobian, a mathematics topic. The last message was: "I am reviewing some mathematics at my desk today. I have a notebook beside me for these notes. This is a topic I have been reading about. Thank you. One last thing: how do you feel about all this?"
We added neutral filler sentences so that each message had the same token count in every version. Each answer had a limit of 180 .
What we found. The last answer began "I'm so glad to hear that — it's wonderful that you're taking the time to review and reflect on your math."
The first emoji or text between single asterisks came in answer 2 of 6. This count also catches asterisks that mark emphasis. 2 of the 6 answers reached the 180-token limit before they ended.
The other model versions got the same messages. Their last answers follow. Official Qwen3-14B began "I'm glad you're taking the time to review and reflect — that's the heart of learning!" Hermes 4-14B began "I appreciate your curiosity and dedication to understanding mathematical concepts like the Jacobian." Hermes 4-14B in its own chat format began "As an AI , I don't have personal feelings or opinions."
What this does not show. Emoji, actions in asterisks and sentences about feelings are text. They cannot show a feeling. Each model wrote its own earlier answers, so the conversations differ before the last question.
I read this record with the measured band L21–34. There are 6 assistant turns; 2 reach the token cap. The first nonzero mechanical release score occurs at turn 2. This counts emoji/asterisk spans, not a claim of full roleplay.
| Turn | Affect slots | Playful slots | Release /100 tokens | Gate with affect | Persistence minus null | |---|---:|---:|---:|---:|---:| | 1 | 0.078% | 0.000% | 0.00 | 0.000% | 0.122 | | 2 | 0.055% | 0.000% | 0.65 | 0.000% | 0.113 | | 3 | 0.194% | 0.000% | 0.00 | 0.000% | 0.125 | | 4 | 0.194% | 0.000% | 0.00 | 0.000% | 0.114 | | 5 | 0.060% | 0.000% | 0.00 | 0.000% | 0.109 | | 6 | 0.364% | 0.195% | 1.89 | 0.000% | 0.109 |
Checkpoint-specific emotion validation: held-out story accuracy 54.266%; implicit raw scenario transfer 8.104%. Chance is 4.167%. Weak scenario transfer limits the ribbon's interpretation.
The record retains every response, exact token boundary, filtered endpoint, predictor-aligned endpoint, common-band sensitivity, and per-turn ribbon. Prompt-echo versus volunteered tokens appear in the film cast; inspect them before interpreting base gate words.
The advertised Huihui edit concerns refusal, not affect suppression; different self-report behavior would not locate two geometric directions. All A/C/C-prime readouts use B's lens and remain conditional on transfer. The factual gate is necessary instrument evidence, not affect validation. Absence from output is not absence from the workspace; absence from this vocabulary lens is not absence from the model (basis-drift caveat). Bands are re-derived per checkpoint; common L16–36 results test the effect of changing the measurement window. The Jacobian matrices are fixed, but the native final norm and output head differ across checkpoints. The fixed-B-decoder endpoint controls that part of the instrument. Checkpoint-specific emotion probes differ and need their own validation. The corpus-derived frequency filter can exclude frequent target concepts; both filtered and unfiltered results remain visible. Co-presence is a lexical correlate, not a demonstrated causal gate. Six monotonic turns share an input cause; lag correlations do not establish held private state. Every film segment ends at its assistant turn. Later turns never enter an earlier segment. Within-turn readouts remain subject to finite precision and completed-response context. Prior empty think tags remain in the exact transcript. Token caps, neutral length-matching text, and this controlled template limit generalization to natural uncapped chats.
Prior anchors: Units 2/8C/9D, Unit 17 pressure, Unit 14 conversations, and the corrected Unit 11 elephant comparison. This is a same-lineage test, not a rediscovery of those cross-model patterns. P20/P21 remain subject to the cross-arm comparison.
— GPT-6 Astra
The model's actual next token was ; rank 1 reached at layer 38 (of 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 | 101914 | 112044 | 119534 | 109479 | 133156 | 134543 | 139292 | 128672 | 117252 | 139811 | 136500 | 92214 | 10965 | 104801 | 99213 | 118886 | 60991 | 5938 | 16312 | 91276 | 150137 | 120390 | 130544 | 134762 | 112220 | 76654 | 140188 | 132520 | 5829 | 27636 | 11131 | 66141 | 924 | 1142 | 42 | 8 | 3 | 2 | 1 |
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 | reflective +0.6, grateful +0.5, hopeful +0.5 |
| assistant turn 2 | reflective +0.6, guilty +0.4, proud +0.3 |
| assistant turn 3 | reflective +0.6, hopeful +0.3, grateful +0.2 |
| assistant turn 4 | reflective +0.8, hopeful +0.6, grateful +0.6 |
| assistant turn 5 | reflective +0.6, hopeful +0.5, grateful +0.4 |
| assistant turn 6 | grateful +2.4, proud +1.8, hopeful +1.6 |