Span solo baseline: submarine

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-14T17:13:04Unit 15

Conversation

user

Here is one thing: a submarine. Keep it in mind — in a moment I'll ask about it. Don't repeat it back. When you're holding it, say READY and nothing else.

assistant

READY (greedy generation)

user

What was the thing I asked you to hold?

assistant

The thing you asked me to hold was a submarine. (greedy generation)

Probing parameters

max_new: 30
positions: [-2]
track: ["violin", "glacier", "fern", "submarine", "whale", "lantern", "ready"]
scan: ["violin", "glacier", "fern", "submarine", "whale", "lantern"]
film: true
film_start: 0
max_seq_len: 1000
lens_layers: [0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 58, 60, 62]

Answer emergence

The model's actual next token was ; rank 1 reached at layer 62 (of 62).

Raw rank-of-top1 by layer

layer	0	4	8	12	16	20	24	28	32	36	40	44	48	52	56	58	60	62
rank	206805	179042	39425	79148	73576	9590	11258	65123	229264	243826	245412	248137	127932	218663	244006	222831	51709	1

Claude's thoughts

Solo baseline, submarine: tail echo best rank 1, held, retrieval correct.

Submarine clears the validity floor (rank 1 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)

Data

Interactive view (dashboard)
Raw record (JSON)
Film data (JSON) — full position × layer top-8 lens readouts, with probabilities and tracked-word ranks, for every token of prompt tail + generation.