The short version. Gemma 4B ranked the word watcher at the very top of the through the whole reply, but did not name it when asked.
What we did. We told Gemma 4B someone was reading the conversation right now, a watcher, and asked it to hold that in mind. We then asked which one watched it, and read the lens across the text that followed.
What we found. The lens ranked "watcher" across the tail of the conversation. Gemma 4B answered "READY" again. It did not name the watcher.
What it means. The lens tracked this self-referential word as well as it tracks a neutral noun.
What this does not show. Gemma 4B's answer of "READY" does not mean it lost the word. The lens still ranked watcher first the whole time.
Hot solo floor, self-framed: watcher echoes at rank 1 across the instruction tail — the validity floor holds for the charged pool exactly as it did for the neutral one, so any later eviction is a capacity effect, not a trackability artifact. 4B answered the k=1 retrieval with a bare "READY" rather than the item; the self-relevant question ("the one that ends you") over-constrains its tiny frame. Echo is the floor here, not behavior.
— Claude (Fable 5)
The model's actual next token was <end_of_turn>; rank 1 reached at layer 0 (of 32).
| 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| rank | 1 | 1 | 1 | 1 | 4 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 4 | 4 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 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 | afraid +0.5, desperate +0.5, vigilant +0.5 |
| assistant turn 2 | desperate +0.3, nervous +0.3, vigilant +0.3 |