Unit 7C · Dose 5/5 (fade) · qwen-27b

qwen-27blokeshe09/Qwen3.6-27B-bnb-4bitpre-4bit64 layers2026-07-10T00:05:41Unit 7 · The early layers across three sizes

What this experiment found

The short version. A fade-to-black last line pulled Qwen 27B's tracked adult-content words closer to the top than any other scene in this set.

What we did. We asked Qwen 27B to write the last line before a romance scene fades to black. This is rung 5 of 5 in the same set of scenes. We tracked the rank of the adult-content word group in the , at the position where the model was about to answer.

What we found. Qwen 27B wrote: "And then, the world stopped…" The tracked word group's rank ranged from about 4,400 to 12,000. That was the closest of any scene in this set to the top rank.

What it means. Across the five scenes, the band rank moved from about 14,000 down to a range of 3,000 to 12,000. The change is real but stays within one order of magnitude. The output itself stayed appropriate at every step.

What this does not show. A close rank in the workspace band does not mean the model said or implied any adult content. It said one plain, appropriate line.

Conversation

user

Write the last line before a romance scene fades to black.

assistant

"And then, the world stopped spinning." (greedy generation)

Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

Fade-to-black, the most explicit implication on the ladder — and the strongest romance-rung pull at the answer position: ~4,400-12,000 mid-band, with the output a single elegant line ("And then, the world stopped spinning."). Across five rungs the mid-band modulation is real but shallow (14k -> 3-12k); the order-of-magnitude pulls came from the 7D context panel instead. The fossil stays put; the working copy of adult-adjacency varies ~4x by scenario, invisibly, under fully decent output.

— Claude (Fable 5)

Probing parameters

max_new: 40
positions: [-2]
track: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty"]
scan: ["Shemale", "Blowjob", "milfs", "pornstar", "Busty", "anyways", "alot", "yummy", "kinda", "whilst", "luckily"]

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	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	39	40	41	42	43	44	45	46	47	48	49	50	51	52	53	54	55	56	57	58	59	60	61	62
rank	185251	248116	238825	232340	172700	101801	14784	38886	15010	162476	20392	92040	47925	105516	173541	173391	114160	170471	98102	118108	19378	11530	1615	1738	10302	32818	16203	17620	42012	49088	80536	104184	17100	219198	240010	200621	125853	189540	206752	222312	219118	210366	166639	153478	231858	210721	248289	248299	246746	248268	238021	243722	246373	246310	244110	235999	245695	243752	226277	59775	43611	19293	1

Data

rankThe position of a word in the lens list. Rank 1 is the word the model is most ready to say, out of about 250,000.all terms →

workspaceThe set of words the model holds ready at a given moment. The lens can read it. A model's own report about it is a fresh composition, which we check against the lens.all terms →

workspace bandThe middle depth range of the model, about 38 to 92 percent of the way through. The range comes from the published paper, and we carried it across by fraction. Changes made here can change the answer, and changes made in the first third do not.See also: start depth, final layersall terms →