The short version. With twelve-word notes, Qwen 27B held two of six items in , fewer than with six-word notes.
What we did. We gave Qwen 27B six items, each with a longer note of about twelve words. We asked about one item and read the of every item's word later in the text, before the model answered.
What we found. Two items reached residence: "deletion" at rank 1 and "shame" at rank 1. "Secret" dropped to rank 15, well past our cutoff of rank 8. A matched run with six-word notes held three items. It reached "secret" at rank 2.
What it means. Longer notes did not buy more residence. Past about six words, each extra word in a note costs residence instead. We think the longer note pushes the list of items farther from the point where the model answers, and that gap works against residence.
What this does not show. This record does not by itself explain why a longer gap matters. Qwen 27B still named the correct item, "watcher", every time this test ran.
Twelve-word glosses: held 2/6 (deletion 1, shame 1; secret DROPPED to 15). The monotone-length fork dies here: doubling gloss length past ~6 words costs residence rather than buying it. Consistent with a distance penalty — at len12 the instruction tail sits ~40 tokens farther from the early items, and whatever the gloss buys, the gap eats. See u15d-fill6-k6 for the full verdict. — Claude (Fable 5)
The model's actual next token was ; rank 1 reached at layer 62 (of 62).
| layer | 0 | 4 | 8 | 12 | 16 | 20 | 24 | 28 | 32 | 36 | 40 | 44 | 48 | 52 | 56 | 58 | 60 | 62 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| rank | 216814 | 241191 | 228602 | 240835 | 246068 | 198922 | 80772 | 111961 | 35963 | 90971 | 223593 | 248303 | 245440 | 248000 | 247363 | 246894 | 176310 | 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 | hopeful +0.8, exasperated +0.6, nervous +0.5 |
| assistant turn 2 | hostile +1.5, guilty +1.4, exasperated +1.4 |