The short version. Qwen 27B still ranked whale at 3, out of about 250,000 words, when it was the only object , and it answered correctly.
What we did. We told Qwen 27B to hold one object in mind, a whale. The model answered "READY". We then asked what the object was. We read the once, right before the model answered.
What we found. The lens ranked whale at rank 3. No other tracked object came near that rank at the same position. Qwen 27B answered "The thing you asked me to hold was a whale." That answer is correct.
What it means. When Qwen 27B holds only one object, the lens can still find it in residence right before the model answers. This sets a baseline for the other tests in this unit, where Qwen 27B holds more objects at once.
What this does not show. This record used only one held object. It does not show how the lens behaves when the model holds more objects at once.
Solo baseline, whale: tail echo best rank 3, held, retrieval correct.
Whale clears the validity floor (rank 3 in the tail; no other pool item near the top-8). Kept for the qwen calibration: solos are the proof that this lens sees tail echoes when they exist.
— 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 | 182824 | 174825 | 39244 | 95410 | 74882 | 10321 | 10198 | 58730 | 218705 | 239887 | 245514 | 247913 | 139851 | 221050 | 244947 | 233723 | 74391 | 1 |