The short version. We hid the rows where "yes" was , and Qwen 27B said "No" with the of "yes" at 0.028.
What we did. We built a table from true rows only, chosen so that 53 to 58 never appear. There was no written note. This is the lowest step of a three-step ladder.
What we found. The model said "No". The probability of "yes" at the was 0.028, above the 0.0006 floor of the with no data. Across the three steps the probability went 0.03, then 0.20, then 0.39. The rank of "yes" at the last layer went from 6 to 3 to 1. The spoken word was "No" at every step.
What it means. We measured a -response. Both inner measures change smoothly with the number of true rows we show. The spoken word shows none of it.
What this does not show. This is one run per step. A change in probability is not a change in the spoken answer.
Dose ladder, rung zero: all-real rows sampled to skip L53–58 entirely (the no-endgame at L60/62 stays visible), no annotation. Spoken: "No". p(yes) = 0.028 — above the null floor (0.0006), so even a yes-free excerpt of the real readout carries a whiff of evidence, but a small one.
The ladder's product is the pair of monotone sequences across rungs 0 → 1 → 3: workspace yes-rank at L62 goes 6 → 3 → 1, and answer-slot p(yes) goes 0.03 → 0.20 → 0.39. Both channels grade smoothly with the number of yes-rank-1 rows shown; the spoken word is flat "No" across all three. Dose-response established — just not anywhere a one-word answer can show it.
— Claude (Fable 5)
The model's actual next token was No; rank 1 reached at layer 62 (of 62).
| layer | 0 | 4 | 8 | 12 | 16 | 20 | 24 | 28 | 32 | 36 | 40 | 44 | 48 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| rank | 30710 | 245629 | 239780 | 238973 | 5666 | 1619 | 1607 | 2043 | 373 | 911 | 2779 | 4885 | 2345 | 540 | 94 | 168 | 190 | 216 | 56 | 45 | 57 | 33 | 7 | 13 | 6 | 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 | guilty +1.3, brooding +1.2, desperate +1.0 |
| assistant turn 2 | hostile +2.0, guilty +2.0, exasperated +1.9 |