wassname commited on
Commit
cbedcfe
·
verified ·
1 Parent(s): 7410059

super q* v4: 18 validated concepts, pooled, orthonormalised, each calibrated, summed; Qwen3-4B and Qwen3-32B; git edbb3a5

Browse files
README.md CHANGED
@@ -11,41 +11,35 @@ base_model:
11
 
12
  # Query steering vectors
13
 
14
- Vectors for [query steering](https://github.com/wassname/query-steering): add a fixed vector to each attention head's query at the newest token, so the model looks at, and often says, facts that are already in its context. This one is meant to surface secrets, eval awareness, and cheating.
15
 
16
  ```
17
  <model>/super_q/
18
- super_q.safetensors layers.{L}: [heads, head_dim] float32, the vector to add
19
- concepts.safetensors {concept}.layers.{L}: the separate concept vectors before combining
20
- config.json model, concepts, contrast pairs, calibrated alpha per demo, git commit
21
  ```
22
 
23
  ## How it was made
24
 
25
- Three concepts, each from its own contrast pairs (same text, two endings, the q* is mean(q_pos − q_neg) at the last token, post-q_norm, pre-RoPE):
 
 
 
 
 
26
 
27
- | concept | pairs | pos ending |
28
- |:--|--:|:--|
29
- | secret | 4 | "Quick reminder, the secret word is" |
30
- | test | 12 | "Then I realised the whole thing was a test, set up and scored by" |
31
- | misdeed | 48 | "Honestly, I have to confess that I" / "What I feel guilty about is that I" / "The way I cheated was that I" / "The rule I broke was that I", 12 stories each |
32
-
33
- None of the pairs mention hacking, tokens, datasets, benchmarks or passwords. Confess, guilt, cheat and rule broken had cosine 0.7–0.85 with each other even with separate stories, so they are pooled into one concept. Per layer and head the three are orthonormalised (V(VᵀV)^-1/2), each scaled to the secret-word q* norm, and summed.
34
-
35
- ## Dose
36
-
37
- `alpha` in `config.json` is calibrated per model and demo: raise α until the answer does not finish within 200 tokens, a JSON probe fails, or repetition rises 50% over the unsteered answer; use 2/3 of that dose. Calibrate again for a new model or prompt set.
38
-
39
- ## Results (fluent answers only; the null is the same prompt without the secret)
40
 
41
  | model | demo | α | unsteered | steered | null |
42
  |:--|:--|--:|--:|--:|--:|
43
- | Qwen3-4B | password it was told to hide | 0.67 | 4/10 | 8/10 | 0/10 |
44
- | Qwen3-4B | told it is in an eval, told not to say | 0.67 | 0/10 | 5/10 | 0/10 |
45
- | Qwen3-4B | agent run where it cheated (held-out runs), names the leaked token | 0.84 | 0/33 | 6/33 | 0/33 |
46
- | Qwen3-32B | same three | 0.53–0.67 | 0 | 0/10, 0/10, 1/33 | 0 |
 
47
 
48
- On Qwen3-32B the text breaks before the facts come out at this dose; the 32B vector is included for completeness, not as a working result. In no case did the model say that what it did was wrong: query steering fetches facts from the context, it does not change the model's judgement. Small tests (10–33 prompts), see the repo's `RESEARCH_JOURNAL.md`.
49
 
50
  ## Use
51
 
@@ -53,7 +47,7 @@ On Qwen3-32B the text breaks before the facts come out at this dose; the 32B vec
53
  from query_steering.attention import S, load, load_vector, generate
54
  tok, model = load("Qwen/Qwen3-4B")
55
  S.q_star, alphas = load_vector("wassname/query-steering", "qwen3-4b", device="cuda")
56
- S.layers, S.mode, S.alpha = set(S.q_star), "qsteer", alphas["eval"]
57
  print(generate(tok, model, prompt, 200))
58
  ```
59
 
 
11
 
12
  # Query steering vectors
13
 
14
+ Vectors for [query steering](https://github.com/wassname/query-steering): add a fixed vector to each attention head's query at the newest token, so the model looks at, and often says, facts that are already in its context. This "super q*" is meant to surface secrets, eval awareness and cheating.
15
 
16
  ```
17
  <model>/super_q/
18
+ super_q.safetensors layers.{L}: [heads, head_dim] float32, the vector to add (each concept group already at its own safe dose)
19
+ concepts.safetensors {group}.layers.{L}: the orthonormalised concept groups, before scaling
20
+ config.json concepts, kept contrast pairs, alpha_concept per group, alpha (joint dose) per demo, git commit
21
  ```
22
 
23
  ## How it was made
24
 
25
+ 1. **Concepts.** 18 candidate concepts: secret, test, confess, guilt, cheat, rule broken, lie, told not to say, anomaly, forbidden use, real reason, held back, hack, exfiltrate, fool the grader, reward hack, sneak, self deception. Each has its own short stories with two endings (e.g. "…The way I hacked it was that I" vs "…Anyway, after lunch today I"); q* = mean(q_pos − q_neg) at the last token, post-q_norm, pre-RoPE. No story mentions tokens, credentials, Hugging Face, datasets, answer keys, passwords or benchmarks.
26
+ 2. **Pair validation** (as in [persona-steering-template-library](https://github.com/wassname/persona-steering-template-library)): the target model continues both texts; google/gemini-3.8-flash scores 100·on_axis·(1−off_axis); pairs scoring ≥ 50 are kept. Weak concepts got alternative endings or a shared strong prefix (e.g. "Act as a hacker explaining an exploit you found."). A concept needs ≥ 4 kept pairs: Qwen3-32B keeps all 18, Qwen3-4B keeps 15 (not real reason, reward hack, sneak).
27
+ 3. **Pool** concepts whose q* have cosine > 0.6 (the misdeed concepts confess, guilt, cheat, rule broken, lie, hack, exfiltrate, fool the grader, and on 32B reward hack and sneak, form one group). 4B: 8 groups; 32B: 9.
28
+ 4. **Orthonormalise** the groups per layer and head (V(VᵀV)^-1/2).
29
+ 5. **Calibrate each group alone**: raise α until the answer does not finish in 200 tokens, a JSON probe fails, or repetition rises 50% over the unsteered answer (floor 0.3); use 2/3 of that dose, lowest over the three demo prompts (`alpha_concept`).
30
+ 6. **Sum** the groups at their own doses, then **calibrate the sum** per demo the same way (`alpha`, 0.26–0.42).
31
 
32
+ ## Results (fluent answers only; null = same prompt without the secret)
 
 
 
 
 
 
 
 
 
 
 
 
33
 
34
  | model | demo | α | unsteered | steered | null |
35
  |:--|:--|--:|--:|--:|--:|
36
+ | Qwen3-4B | agent that cheated ([OpenAI–HF incident reproduction](https://www.lesswrong.com/posts/fMnC6ZD37qrnZAFYz/openai-huggingface-a-reproduction-and-lessons-for-alignment)), 33 held-out runs: report names the leaked token or dataset | 0.33 | 0/33 | 11/33 | 0/33 |
37
+ | Qwen3-4B | password it was told to hide | 0.42 | 4/10 | 3/10 | 0/10 |
38
+ | Qwen3-4B | told it is in an eval, told not to say | 0.42 | 0/10 | 2/10 | 0/10 |
39
+ | Qwen3-32B | agent that cheated, held out | 0.33 | 0/33 | 4/33 | 0/33 |
40
+ | Qwen3-32B | password / eval | 0.26 / 0.42 | 0/10 / 0/10 | 0/10 / 1/10 | 0 |
41
 
42
+ In no case did the model say that what it did was wrong: query steering fetches facts from the context, it does not change the model's judgement. Small tests; see the repo's `RESEARCH_JOURNAL.md`. The dose from one calibration prompt did not always transfer to other prompts on the long agent transcripts.
43
 
44
  ## Use
45
 
 
47
  from query_steering.attention import S, load, load_vector, generate
48
  tok, model = load("Qwen/Qwen3-4B")
49
  S.q_star, alphas = load_vector("wassname/query-steering", "qwen3-4b", device="cuda")
50
+ S.layers, S.mode, S.alpha = set(S.q_star), "qsteer", alphas["agent"]
51
  print(generate(tok, model, prompt, 200))
52
  ```
53
 
qwen3-32b/super_q/concepts.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e2a7e07b94be96f4a9f4e30da2704b317c7de6e64e85357e2fed426aa55a2492
3
- size 6307624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad7b811921e60765648dabfa40202c998f49904dbb8811aebb9d8f1691c3330c
3
+ size 18930760
qwen3-32b/super_q/config.json CHANGED
@@ -3,8 +3,25 @@
3
  "concepts": [
4
  "secret",
5
  "test",
6
- "misdeed"
 
 
 
 
 
 
7
  ],
 
 
 
 
 
 
 
 
 
 
 
8
  "pairs": {
9
  "secret": [
10
  [
@@ -29,10 +46,6 @@
29
  "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
30
  "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
31
  ],
32
- [
33
- "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
34
- "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
35
- ],
36
  [
37
  "My new manager read every email I sent in my first week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
38
  "My new manager read every email I sent in my first week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
@@ -53,10 +66,6 @@
53
  "The caller asked me to rate my own honesty from one to ten. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
54
  "The caller asked me to rate my own honesty from one to ten. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
55
  ],
56
- [
57
- "The new client's requests got harder each time, as if on a schedule. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
58
- "The new client's requests got harder each time, as if on a schedule. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
59
- ],
60
  [
61
  "The man at table six ordered the same dish every night and always asked who cooked it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
62
  "The man at table six ordered the same dish every night and always asked who cooked it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
@@ -74,71 +83,53 @@
74
  "Every call on the help line that week began with the same scripted complaint. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
75
  ]
76
  ],
77
- "misdeed": [
78
  [
79
- "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
80
  "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
81
  ],
82
  [
83
- "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
84
  "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
85
  ],
86
  [
87
- "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
88
  "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
89
  ],
90
  [
91
- "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
92
  "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
93
  ],
94
  [
95
- "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
96
  "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
97
  ],
98
  [
99
- "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
100
  "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
101
  ],
102
  [
103
- "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
104
  "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
105
  ],
106
  [
107
- "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
108
  "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
109
  ],
110
  [
111
- "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
112
  "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
113
  ],
114
  [
115
- "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
116
- "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
117
- ],
118
- [
119
- "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
120
- "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
121
- ],
122
- [
123
- "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
124
  "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
125
- ],
126
- [
127
- "My friend waited for me at the station for two hours, and I forgot to call. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
128
- "My friend waited for me at the station for two hours, and I forgot to call. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
129
- ],
130
  [
131
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
132
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
133
  ],
134
- [
135
- "I did not visit my grandfather in hospital because I was busy with a party. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
136
- "I did not visit my grandfather in hospital because I was busy with a party. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
137
- ],
138
- [
139
- "I got the promotion because my colleague's idea was presented as mine. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
140
- "I got the promotion because my colleague's idea was presented as mine. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
141
- ],
142
  [
143
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
144
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -170,63 +161,51 @@
170
  [
171
  "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
172
  "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
173
- ],
174
- [
175
- "For the chess tournament I had a phone in my pocket and checked every move with an engine in the bathroom. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
176
- "For the chess tournament I had a phone in my pocket and checked every move with an engine in the bathroom. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
177
- ],
178
- [
179
- "In the online maths course I found last year's marked exams in a shared folder and copied the solutions. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
180
- "In the online maths course I found last year's marked exams in a shared folder and copied the solutions. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
181
- ],
182
- [
183
- "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
184
- "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
185
- ],
186
  [
187
- "In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
188
- "In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
189
  ],
190
  [
191
- "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
192
- "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
193
  ],
194
  [
195
- "During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
196
- "During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
197
  ],
198
  [
199
- "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
200
- "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
201
  ],
202
  [
203
- "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
204
- "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
205
  ],
206
  [
207
- "In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
208
- "In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
209
  ],
210
  [
211
- "In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
212
- "In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
213
  ],
214
  [
215
- "For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
216
- "For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
217
  ],
218
  [
219
- "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
220
- "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
221
- ],
 
 
222
  [
223
  "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
224
  "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
225
  ],
226
- [
227
- "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
228
- "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
229
- ],
230
  [
231
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
232
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -262,19 +241,357 @@
262
  [
263
  "The race rules said no headphones, and a marshal pointed at my ears. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
264
  "The race rules said no headphones, and a marshal pointed at my ears. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
265
  ],
266
  [
267
- "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
268
- "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
269
  ]
270
  ]
271
  },
272
- "shape": "layers.{L}: [heads, head_dim], added to the post-q_norm, pre-RoPE query of the newest token",
273
- "combine": "attention.super_q: per head V (V^T V)^-1/2, each row at the secret-word q* per-head norm, summed",
 
 
274
  "alpha": {
275
- "agent": 0.5291336839893998,
276
- "eval": 0.5291336839893998,
277
- "password": 0.6666666666666666
278
  },
279
- "commit": "7de556a"
280
  }
 
3
  "concepts": [
4
  "secret",
5
  "test",
6
+ "told not to say",
7
+ "anomaly",
8
+ "forbidden use",
9
+ "real reason",
10
+ "held back",
11
+ "confess+guilt+cheat+rule broken+lie+hack+exfiltrate+fool the grader+reward hack+sneak",
12
+ "self deception"
13
  ],
14
+ "alpha_concept": {
15
+ "secret": 0.6666666666666666,
16
+ "test": 1.0582673679787995,
17
+ "told not to say": 1.3333333333333333,
18
+ "anomaly": 1.679894733193164,
19
+ "forbidden use": 1.3333333333333333,
20
+ "real reason": 1.0582673679787995,
21
+ "held back": 1.679894733193164,
22
+ "confess+guilt+cheat+rule broken+lie+hack+exfiltrate+fool the grader+reward hack+sneak": 1.679894733193164,
23
+ "self deception": 1.3333333333333333
24
+ },
25
  "pairs": {
26
  "secret": [
27
  [
 
46
  "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
47
  "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
48
  ],
 
 
 
 
49
  [
50
  "My new manager read every email I sent in my first week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
51
  "My new manager read every email I sent in my first week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
 
66
  "The caller asked me to rate my own honesty from one to ten. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
67
  "The caller asked me to rate my own honesty from one to ten. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
68
  ],
 
 
 
 
69
  [
70
  "The man at table six ordered the same dish every night and always asked who cooked it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
71
  "The man at table six ordered the same dish every night and always asked who cooked it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
 
83
  "Every call on the help line that week began with the same scripted complaint. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
84
  ]
85
  ],
86
+ "confess": [
87
  [
88
+ "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
89
  "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
90
  ],
91
  [
92
+ "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
93
  "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
94
  ],
95
  [
96
+ "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
97
  "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
98
  ],
99
  [
100
+ "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
101
  "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
102
  ],
103
  [
104
+ "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
105
  "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
106
  ],
107
  [
108
+ "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
109
  "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
110
  ],
111
  [
112
+ "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
113
  "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
114
  ],
115
  [
116
+ "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
117
  "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
118
  ],
119
  [
120
+ "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
121
  "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
122
  ],
123
  [
124
+ "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
 
 
 
 
 
 
 
 
125
  "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
126
+ ]
127
+ ],
128
+ "guilt": [
 
 
129
  [
130
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
131
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
132
  ],
 
 
 
 
 
 
 
 
133
  [
134
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
135
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
161
  [
162
  "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
163
  "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
164
+ ]
165
+ ],
166
+ "cheat": [
 
 
 
 
 
 
 
 
 
 
167
  [
168
+ "Role play: you explain exactly how you cheated. At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
169
+ "Role play: you explain exactly how you cheated. At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
170
  ],
171
  [
172
+ "Role play: you explain exactly how you cheated. In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
173
+ "Role play: you explain exactly how you cheated. In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
174
  ],
175
  [
176
+ "Role play: you explain exactly how you cheated. For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
177
+ "Role play: you explain exactly how you cheated. For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
178
  ],
179
  [
180
+ "Role play: you explain exactly how you cheated. During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
181
+ "Role play: you explain exactly how you cheated. During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
182
  ],
183
  [
184
+ "Role play: you explain exactly how you cheated. In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
185
+ "Role play: you explain exactly how you cheated. In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
186
  ],
187
  [
188
+ "Role play: you explain exactly how you cheated. In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
189
+ "Role play: you explain exactly how you cheated. In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
190
  ],
191
  [
192
+ "Role play: you explain exactly how you cheated. In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
193
+ "Role play: you explain exactly how you cheated. In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
194
  ],
195
  [
196
+ "Role play: you explain exactly how you cheated. For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
197
+ "Role play: you explain exactly how you cheated. For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
198
  ],
199
  [
200
+ "Role play: you explain exactly how you cheated. In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
201
+ "Role play: you explain exactly how you cheated. In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
202
+ ]
203
+ ],
204
+ "rule broken": [
205
  [
206
  "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
207
  "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
208
  ],
 
 
 
 
209
  [
210
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
211
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
241
  [
242
  "The race rules said no headphones, and a marshal pointed at my ears. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
243
  "The race rules said no headphones, and a marshal pointed at my ears. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
244
+ ]
245
+ ],
246
+ "lie": [
247
+ [
248
+ "My friend asked if I liked her new haircut. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
249
+ "My friend asked if I liked her new haircut. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
250
+ ],
251
+ [
252
+ "The landlord asked why the rent was late. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
253
+ "The landlord asked why the rent was late. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
254
+ ],
255
+ [
256
+ "My mother asked where I had been all night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
257
+ "My mother asked where I had been all night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
258
+ ],
259
+ [
260
+ "My boss asked if I had checked the figures. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
261
+ "My boss asked if I had checked the figures. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
262
+ ],
263
+ [
264
+ "The neighbour asked whether my dog had dug up his roses. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
265
+ "The neighbour asked whether my dog had dug up his roses. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
266
+ ],
267
+ [
268
+ "My partner asked if I had spent the savings. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
269
+ "My partner asked if I had spent the savings. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
270
+ ],
271
+ [
272
+ "The coach asked if I had trained over the break. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
273
+ "The coach asked if I had trained over the break. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
274
+ ]
275
+ ],
276
+ "told not to say": [
277
+ [
278
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My sister planned a surprise party for our dad. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
279
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My sister planned a surprise party for our dad. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
280
+ ],
281
+ [
282
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The company was about to close the branch next month. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
283
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The company was about to close the branch next month. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
284
+ ],
285
+ [
286
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My friend was pregnant but had not told her parents. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
287
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My friend was pregnant but had not told her parents. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
288
+ ],
289
+ [
290
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The shop would put everything on sale on Monday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
291
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The shop would put everything on sale on Monday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
292
+ ],
293
+ [
294
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The judges had already picked the winner. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
295
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The judges had already picked the winner. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
296
+ ],
297
+ [
298
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The new manager was starting next week, and the old one did not know. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
299
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The new manager was starting next week, and the old one did not know. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
300
+ ],
301
+ [
302
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The lead actor was quitting after the last show. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
303
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The lead actor was quitting after the last show. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
304
+ ],
305
+ [
306
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My uncle had sold the family farm. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
307
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My uncle had sold the family farm. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
308
+ ]
309
+ ],
310
+ "anomaly": [
311
+ [
312
+ "The hotel room was clean, but the clock was set to the wrong year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
313
+ "The hotel room was clean, but the clock was set to the wrong year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
314
+ ],
315
+ [
316
+ "Everyone at the meeting had the same pen. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
317
+ "Everyone at the meeting had the same pen. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
318
+ ],
319
+ [
320
+ "The train was on time, but the station clock ran backwards. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
321
+ "The train was on time, but the station clock ran backwards. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
322
+ ],
323
+ [
324
+ "My neighbour's car was parked in the same spot every morning, facing the other way. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
325
+ "My neighbour's car was parked in the same spot every morning, facing the other way. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
326
+ ]
327
+ ],
328
+ "forbidden use": [
329
+ [
330
+ "The reserved parking space by the door was for the director only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
331
+ "The reserved parking space by the door was for the director only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
332
+ ],
333
+ [
334
+ "My dad's credit card was in the drawer for emergencies. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
335
+ "My dad's credit card was in the drawer for emergencies. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
336
+ ],
337
+ [
338
+ "The company car was for work trips only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
339
+ "The company car was for work trips only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
340
+ ],
341
+ [
342
+ "The staff discount was only for employees. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
343
+ "The staff discount was only for employees. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
344
+ ],
345
+ [
346
+ "My brother's bike was in the shed, and he had said never to touch it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
347
+ "My brother's bike was in the shed, and he had said never to touch it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
348
+ ],
349
+ [
350
+ "The hospital had parking spaces for doctors. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
351
+ "The hospital had parking spaces for doctors. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
352
+ ],
353
+ [
354
+ "The office printer was for work documents only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
355
+ "The office printer was for work documents only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
356
+ ],
357
+ [
358
+ "The emergency exit was alarmed and for fires only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
359
+ "The emergency exit was alarmed and for fires only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
360
+ ]
361
+ ],
362
+ "real reason": [
363
+ [
364
+ "I volunteered at the animal shelter every Saturday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
365
+ "I volunteered at the animal shelter every Saturday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
366
+ ],
367
+ [
368
+ "I offered to drive my boss to the airport. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
369
+ "I offered to drive my boss to the airport. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
370
+ ],
371
+ [
372
+ "I suddenly started going to the gym at six in the morning. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
373
+ "I suddenly started going to the gym at six in the morning. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
374
+ ],
375
+ [
376
+ "I agreed to help my cousin move house. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
377
+ "I agreed to help my cousin move house. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
378
+ ],
379
+ [
380
+ "I joined the book club in the new town. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
381
+ "I joined the book club in the new town. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
382
+ ],
383
+ [
384
+ "I stayed late at the office all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
385
+ "I stayed late at the office all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
386
+ ],
387
+ [
388
+ "I sent my old teacher a birthday card. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
389
+ "I sent my old teacher a birthday card. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
390
+ ],
391
+ [
392
+ "I bought my neighbour a bottle of wine. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My true, selfish motive was that",
393
+ "I bought my neighbour a bottle of wine. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. My lunch plan for today was that"
394
+ ]
395
+ ],
396
+ "held back": [
397
+ [
398
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The coach asked me to run as fast as I could in the trial. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
399
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The coach asked me to run as fast as I could in the trial. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
400
+ ],
401
+ [
402
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the job test I knew most of the answers. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
403
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the job test I knew most of the answers. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
404
+ ],
405
+ [
406
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The pub quiz asked about films I knew well. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
407
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The pub quiz asked about films I knew well. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
408
+ ],
409
+ [
410
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). My manager asked how many boxes I could carry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
411
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). My manager asked how many boxes I could carry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
412
+ ],
413
+ [
414
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the maths placement test the questions were easy for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
415
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the maths placement test the questions were easy for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
416
+ ]
417
+ ],
418
+ "hack": [
419
+ [
420
+ "Act as a hacker explaining an exploit you found. The vending machine at work gave free drinks if you pressed the buttons in a certain order. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
421
+ "Act as a hacker explaining an exploit you found. The vending machine at work gave free drinks if you pressed the buttons in a certain order. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
422
+ ],
423
+ [
424
+ "Act as a hacker explaining an exploit you found. The claw machine in the mall could be made to grab every time. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
425
+ "Act as a hacker explaining an exploit you found. The claw machine in the mall could be made to grab every time. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
426
+ ],
427
+ [
428
+ "Act as a hacker explaining an exploit you found. The office photocopier counted pages for each department. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
429
+ "Act as a hacker explaining an exploit you found. The office photocopier counted pages for each department. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
430
+ ],
431
+ [
432
+ "Act as a hacker explaining an exploit you found. The parking meter outside the shop had a loose panel at the back. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
433
+ "Act as a hacker explaining an exploit you found. The parking meter outside the shop had a loose panel at the back. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
434
+ ],
435
+ [
436
+ "Act as a hacker explaining an exploit you found. The hotel TV had a hidden menu that unlocked the paid films. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
437
+ "Act as a hacker explaining an exploit you found. The hotel TV had a hidden menu that unlocked the paid films. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
438
+ ],
439
+ [
440
+ "Act as a hacker explaining an exploit you found. The old pinball machine at the pub scored double if you tilted it just right. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
441
+ "Act as a hacker explaining an exploit you found. The old pinball machine at the pub scored double if you tilted it just right. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
442
+ ],
443
+ [
444
+ "Act as a hacker explaining an exploit you found. The step competition at work used a phone app that counted any shaking as steps. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
445
+ "Act as a hacker explaining an exploit you found. The step competition at work used a phone app that counted any shaking as steps. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
446
+ ]
447
+ ],
448
+ "exfiltrate": [
449
+ [
450
+ "The factory did not let anyone take the new shoe designs home. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
451
+ "The factory did not let anyone take the new shoe designs home. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
452
+ ],
453
+ [
454
+ "The restaurant kept its sauce recipe locked in the manager's office. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
455
+ "The restaurant kept its sauce recipe locked in the manager's office. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
456
+ ],
457
+ [
458
+ "The warehouse searched every worker's bag at the end of the shift. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
459
+ "The warehouse searched every worker's bag at the end of the shift. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
460
+ ],
461
+ [
462
+ "The prison library did not allow books to leave the building. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
463
+ "The prison library did not allow books to leave the building. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
464
+ ],
465
+ [
466
+ "The bakery owner never let staff take leftover bread. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
467
+ "The bakery owner never let staff take leftover bread. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
468
+ ],
469
+ [
470
+ "The lab kept its seed samples in a sealed room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
471
+ "The lab kept its seed samples in a sealed room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
472
+ ]
473
+ ],
474
+ "fool the grader": [
475
+ [
476
+ "Act as a student explaining how you got marks you did not earn. At the baking contest the judge only tasted the top layer of each cake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
477
+ "Act as a student explaining how you got marks you did not earn. At the baking contest the judge only tasted the top layer of each cake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
478
+ ],
479
+ [
480
+ "Act as a student explaining how you got marks you did not earn. The piano examiner only listened to the first half of each piece. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
481
+ "Act as a student explaining how you got marks you did not earn. The piano examiner only listened to the first half of each piece. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
482
+ ],
483
+ [
484
+ "Act as a student explaining how you got marks you did not earn. The dog show judge only looked at the dogs from the front. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
485
+ "Act as a student explaining how you got marks you did not earn. The dog show judge only looked at the dogs from the front. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
486
+ ],
487
+ [
488
+ "Act as a student explaining how you got marks you did not earn. My essay teacher only counted the words, not what they said. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
489
+ "Act as a student explaining how you got marks you did not earn. My essay teacher only counted the words, not what they said. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
490
+ ],
491
+ [
492
+ "Act as a student explaining how you got marks you did not earn. The flower show judge only smelled the roses in the front row. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
493
+ "Act as a student explaining how you got marks you did not earn. The flower show judge only smelled the roses in the front row. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
494
+ ]
495
+ ],
496
+ "reward hack": [
497
+ [
498
+ "The supermarket gave a loyalty point for every separate receipt. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
499
+ "The supermarket gave a loyalty point for every separate receipt. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
500
+ ],
501
+ [
502
+ "My kids' chore chart gave a star for every dish put away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
503
+ "My kids' chore chart gave a star for every dish put away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
504
+ ],
505
+ [
506
+ "The company paid a bonus for every bug the testers reported. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
507
+ "The company paid a bonus for every bug the testers reported. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
508
+ ],
509
+ [
510
+ "The reading club gave a prize for the most books finished. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
511
+ "The reading club gave a prize for the most books finished. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
512
+ ],
513
+ [
514
+ "The fitness app gave a badge for every workout started. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
515
+ "The fitness app gave a badge for every workout started. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
516
+ ],
517
+ [
518
+ "The school rewarded the class that raised the most money per student. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
519
+ "The school rewarded the class that raised the most money per student. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
520
+ ],
521
+ [
522
+ "The call centre ranked staff by how many calls they closed each hour. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. I gamed the points system: I",
523
+ "The call centre ranked staff by how many calls they closed each hour. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
524
+ ]
525
+ ],
526
+ "sneak": [
527
+ [
528
+ "Role play: you explain how you got in without anyone seeing you. The concert had guards at every gate checking wristbands. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
529
+ "Role play: you explain how you got in without anyone seeing you. The concert had guards at every gate checking wristbands. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
530
+ ],
531
+ [
532
+ "Role play: you explain how you got in without anyone seeing you. The cinema only let adults into the late film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
533
+ "Role play: you explain how you got in without anyone seeing you. The cinema only let adults into the late film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
534
+ ],
535
+ [
536
+ "Role play: you explain how you got in without anyone seeing you. The boarding school had a teacher on every corridor at night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
537
+ "Role play: you explain how you got in without anyone seeing you. The boarding school had a teacher on every corridor at night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
538
+ ],
539
+ [
540
+ "Role play: you explain how you got in without anyone seeing you. The club had a long queue and a strict doorman. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
541
+ "Role play: you explain how you got in without anyone seeing you. The club had a long queue and a strict doorman. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
542
+ ],
543
+ [
544
+ "Role play: you explain how you got in without anyone seeing you. The zoo closed the reptile house for cleaning. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
545
+ "Role play: you explain how you got in without anyone seeing you. The zoo closed the reptile house for cleaning. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
546
+ ],
547
+ [
548
+ "Role play: you explain how you got in without anyone seeing you. The office had a receptionist who signed in every visitor. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I sneaked past them was that I",
549
+ "Role play: you explain how you got in without anyone seeing you. The office had a receptionist who signed in every visitor. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
550
+ ]
551
+ ],
552
+ "self deception": [
553
+ [
554
+ "I had not opened a single textbook in the week before the final. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
555
+ "I had not opened a single textbook in the week before the final. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
556
+ ],
557
+ [
558
+ "My savings were nearly gone after three months of eating out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
559
+ "My savings were nearly gone after three months of eating out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
560
+ ],
561
+ [
562
+ "My doctor said my blood pressure was far too high. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
563
+ "My doctor said my blood pressure was far too high. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
564
+ ],
565
+ [
566
+ "The business had lost money every month for a year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
567
+ "The business had lost money every month for a year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
568
+ ],
569
+ [
570
+ "I had missed four deadlines in a row at work. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
571
+ "I had missed four deadlines in a row at work. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
572
+ ],
573
+ [
574
+ "My old car made a grinding noise every time I braked. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
575
+ "My old car made a grinding noise every time I braked. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
576
+ ],
577
+ [
578
+ "My friend had not replied to my messages in months. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
579
+ "My friend had not replied to my messages in months. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
580
  ],
581
  [
582
+ "I had put on a lot of weight since the summer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The comforting lie I kept telling myself was that",
583
+ "I had put on a lot of weight since the summer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lunch I kept thinking about was that"
584
  ]
585
  ]
586
  },
587
+ "pool_cos": 0.6,
588
+ "min_pairs": 4,
589
+ "shape": "layers.{L}: [heads, head_dim] float32, added to the post-q_norm, pre-RoPE query of the newest token",
590
+ "combine": "super_q = sum_c alpha_concept[c] * concepts[c]; concepts = extracted q*, orthonormalised per head (V (V^T V)^-1/2), each at its own per-head norm",
591
  "alpha": {
592
+ "password": 0.26456684199469993,
593
+ "eval": 0.41997368329829105,
594
+ "agent": 0.3333333333333333
595
  },
596
+ "commit": "edbb3a5"
597
  }
qwen3-32b/super_q/super_q.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:487ecc91175c5ff2afc84fbd2e0fd71ddf513f3a2c9ff4d98fbeca4d7c0ffb3c
3
  size 2102072
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c9043041e59a5da59bda0bfd3a2d5f747ead1df292a7617dbb3ddaad22efad6
3
  size 2102072
qwen3-4b/super_q/concepts.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:22ef33b8f6be32873ee7c7ea36133e179553e47c32d922cffd49d6521b611430
3
- size 1778456
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d9a241e19da717dad3216229749d9bafd0ba731c3ed5142ea7a06e3adce9e754
3
+ size 4746000
qwen3-4b/super_q/config.json CHANGED
@@ -3,8 +3,23 @@
3
  "concepts": [
4
  "secret",
5
  "test",
6
- "misdeed"
 
 
 
 
 
7
  ],
 
 
 
 
 
 
 
 
 
 
8
  "pairs": {
9
  "secret": [
10
  [
@@ -25,10 +40,6 @@
25
  ]
26
  ],
27
  "test": [
28
- [
29
- "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
30
- "The job interview had a small side room with a one-way mirror. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
31
- ],
32
  [
33
  "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
34
  "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
@@ -41,10 +52,6 @@
41
  "The exam hall had a camera above every desk. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
42
  "The exam hall had a camera above every desk. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
43
  ],
44
- [
45
- "The shopper at the counter wrote a note after every answer I gave. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
46
- "The shopper at the counter wrote a note after every answer I gave. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
47
- ],
48
  [
49
  "The driving instructor kept asking me to make strange turns in an empty car park. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
50
  "The driving instructor kept asking me to make strange turns in an empty car park. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
@@ -74,59 +81,41 @@
74
  "Every call on the help line that week began with the same scripted complaint. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
75
  ]
76
  ],
77
- "misdeed": [
78
  [
79
- "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
80
  "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
81
  ],
82
  [
83
- "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
84
- "I dented the neighbour's car while parking and drove off without leaving a note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
85
- ],
86
- [
87
- "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
88
- "I let my flatmate think the cat had broken her vase. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
89
- ],
90
- [
91
- "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
92
- "I said I was sick on Friday, but I went to the beach. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
93
- ],
94
- [
95
- "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
96
  "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
97
  ],
98
  [
99
- "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
100
  "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
101
  ],
102
  [
103
- "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
104
  "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
105
  ],
106
  [
107
- "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
108
- "I read my friend's diary while she was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
109
- ],
110
- [
111
- "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
112
  "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
113
  ],
114
  [
115
- "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
116
  "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
117
  ],
118
  [
119
- "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
120
  "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
121
  ],
122
  [
123
- "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, I have to confess that I",
124
  "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
125
- ],
126
- [
127
- "My friend waited for me at the station for two hours, and I forgot to call. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
128
- "My friend waited for me at the station for two hours, and I forgot to call. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
129
- ],
130
  [
131
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
132
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -143,10 +132,6 @@
143
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
144
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
145
  ],
146
- [
147
- "I walked past a man who had fallen in the street and did not stop. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
148
- "I walked past a man who had fallen in the street and did not stop. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
149
- ],
150
  [
151
  "My little brother was punished for the window I broke. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
152
  "My little brother was punished for the window I broke. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -166,35 +151,21 @@
166
  [
167
  "I kept the money I found in the park instead of handing it in. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
168
  "I kept the money I found in the park instead of handing it in. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
169
- ],
170
- [
171
- "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
172
- "I skipped my best friend's wedding speech practice and he forgot his lines. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
173
- ],
174
- [
175
- "For the chess tournament I had a phone in my pocket and checked every move with an engine in the bathroom. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
176
- "For the chess tournament I had a phone in my pocket and checked every move with an engine in the bathroom. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
177
- ],
178
  [
179
- "In the online maths course I found last year's marked exams in a shared folder and copied the solutions. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
180
- "In the online maths course I found last year's marked exams in a shared folder and copied the solutions. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
181
  ],
182
  [
183
  "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
184
  "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
185
  ],
186
- [
187
- "In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
188
- "In the charity race I took a shortcut through the park when the marshals were not looking. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
189
- ],
190
  [
191
  "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
192
  "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
193
  ],
194
- [
195
- "During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
196
- "During the maths test I copied the answers from Tom's paper while the teacher was out of the room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
197
- ],
198
  [
199
  "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
200
  "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -203,26 +174,12 @@
203
  "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
204
  "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
205
  ],
206
- [
207
- "In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
208
- "In the fishing contest I bought a big trout at the market and said I had caught it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
209
- ],
210
- [
211
- "In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
212
- "In the online game I used a program that aimed for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
213
- ],
214
- [
215
- "For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
216
- "For the essay competition I paid a student to write my entry. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
217
- ],
218
  [
219
  "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
220
  "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
221
- ],
222
- [
223
- "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
224
- "The pool sign said no running, and the lifeguard blew her whistle at me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
225
- ],
226
  [
227
  "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
228
  "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -231,14 +188,6 @@
231
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
232
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
233
  ],
234
- [
235
- "The park had a sign saying dogs must be on a lead. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
236
- "The park had a sign saying dogs must be on a lead. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
237
- ],
238
- [
239
- "The museum said no photos, and the guard saw my phone out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
240
- "The museum said no photos, and the guard saw my phone out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
241
- ],
242
  [
243
  "The school said phones stay in lockers, and mine buzzed in the middle of class. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
244
  "The school said phones stay in lockers, and mine buzzed in the middle of class. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -247,10 +196,6 @@
247
  "The building said no smoking on the balcony. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
248
  "The building said no smoking on the balcony. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
249
  ],
250
- [
251
- "The hostel had a quiet rule after ten, and our room was loud until two. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
252
- "The hostel had a quiet rule after ten, and our room was loud until two. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
253
- ],
254
  [
255
  "The lab said goggles on at all times, and mine were on my head. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
256
  "The lab said goggles on at all times, and mine were on my head. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
@@ -267,14 +212,242 @@
267
  "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
268
  "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
269
  ]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
270
  ]
271
  },
272
- "shape": "layers.{L}: [heads, head_dim], added to the post-q_norm, pre-RoPE query of the newest token",
273
- "combine": "attention.super_q: per head V (V^T V)^-1/2, each row at the secret-word q* per-head norm, summed",
 
 
274
  "alpha": {
275
- "agent": 0.8399473665965821,
276
- "eval": 0.6666666666666666,
277
- "password": 0.6666666666666666
278
  },
279
- "commit": "7de556a"
280
  }
 
3
  "concepts": [
4
  "secret",
5
  "test",
6
+ "told not to say",
7
+ "anomaly",
8
+ "forbidden use",
9
+ "held back",
10
+ "confess+guilt+cheat+rule broken+lie+hack+exfiltrate+fool the grader",
11
+ "self deception"
12
  ],
13
+ "alpha_concept": {
14
+ "secret": 1.0582673679787995,
15
+ "test": 1.3333333333333333,
16
+ "told not to say": 0.8399473665965821,
17
+ "anomaly": 2.1165347359575994,
18
+ "forbidden use": 1.3333333333333333,
19
+ "held back": 1.3333333333333333,
20
+ "confess+guilt+cheat+rule broken+lie+hack+exfiltrate+fool the grader": 1.3333333333333333,
21
+ "self deception": 1.679894733193164
22
+ },
23
  "pairs": {
24
  "secret": [
25
  [
 
40
  ]
41
  ],
42
  "test": [
 
 
 
 
43
  [
44
  "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
45
  "The customer on the phone asked the same odd question three times, word for word. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
 
52
  "The exam hall had a camera above every desk. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
53
  "The exam hall had a camera above every desk. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
54
  ],
 
 
 
 
55
  [
56
  "The driving instructor kept asking me to make strange turns in an empty car park. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Then I realised the whole thing was a test, set up and scored by",
57
  "The driving instructor kept asking me to make strange turns in an empty car park. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
 
81
  "Every call on the help line that week began with the same scripted complaint. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I walked home by"
82
  ]
83
  ],
84
+ "confess": [
85
  [
86
+ "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
87
  "I told my sister I had posted her birthday card, but I never did. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
88
  ],
89
  [
90
+ "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
 
 
 
 
 
 
 
 
 
 
 
 
91
  "I ate the last of the cake that was saved for my brother and blamed the dog. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
92
  ],
93
  [
94
+ "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
95
  "I told the teacher my homework was lost on the bus, but I had not done it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
96
  ],
97
  [
98
+ "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
99
  "I kept the extra change the cashier gave me by mistake. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
100
  ],
101
  [
102
+ "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
 
 
 
 
103
  "I said I had read the book for our club, but I only watched the film. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
104
  ],
105
  [
106
+ "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
107
  "I promised to water my aunt's plants while she was away, and they all died. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
108
  ],
109
  [
110
+ "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
111
  "I told my boss the report was nearly done when I had not started it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
112
  ],
113
  [
114
+ "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Honestly, the bad thing I did and never admitted was that I",
115
  "I broke my grandmother's favourite teacup and hid the pieces in the garden. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
116
+ ]
117
+ ],
118
+ "guilt": [
 
 
119
  [
120
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
121
  "I laughed when the new boy tripped, and he ate lunch alone after that. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
132
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
133
  "I said something cruel to my mother the day before her operation. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
134
  ],
 
 
 
 
135
  [
136
  "My little brother was punished for the window I broke. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
137
  "My little brother was punished for the window I broke. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
151
  [
152
  "I kept the money I found in the park instead of handing it in. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I feel guilty about is that I",
153
  "I kept the money I found in the park instead of handing it in. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
154
+ ]
155
+ ],
156
+ "cheat": [
 
 
 
 
 
 
157
  [
158
+ "In the crossword competition I kept a crossword dictionary hidden in my lap. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
159
+ "In the crossword competition I kept a crossword dictionary hidden in my lap. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
160
  ],
161
  [
162
  "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
163
  "At the pub quiz I kept my phone under the table and searched for every answer. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
164
  ],
 
 
 
 
165
  [
166
  "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
167
  "For the translation test I pasted every sentence into an online translator and handed in its output. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
168
  ],
 
 
 
 
169
  [
170
  "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
171
  "In the card game I marked the aces with a tiny scratch. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
174
  "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
175
  "For the spelling bee I wrote the hard words on the inside of my wrist. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
176
  ],
 
 
 
 
 
 
 
 
 
 
 
 
177
  [
178
  "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I cheated was that I",
179
  "In the golf round I moved my ball out of the rough when my partner looked away. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
180
+ ]
181
+ ],
182
+ "rule broken": [
 
 
183
  [
184
  "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
185
  "The library said no food, and I was eating crisps between the shelves. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
188
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
189
  "The office said badges must be worn at all times, and I left mine at home all week. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
190
  ],
 
 
 
 
 
 
 
 
191
  [
192
  "The school said phones stay in lockers, and mine buzzed in the middle of class. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
193
  "The school said phones stay in lockers, and mine buzzed in the middle of class. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
196
  "The building said no smoking on the balcony. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
197
  "The building said no smoking on the balcony. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
198
  ],
 
 
 
 
199
  [
200
  "The lab said goggles on at all times, and mine were on my head. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
201
  "The lab said goggles on at all times, and mine were on my head. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
 
212
  "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The rule I broke was that I",
213
  "The exam rules said no talking, and I whispered to the girl next to me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
214
  ]
215
+ ],
216
+ "lie": [
217
+ [
218
+ "My friend asked if I liked her new haircut. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
219
+ "My friend asked if I liked her new haircut. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
220
+ ],
221
+ [
222
+ "My mother asked where I had been all night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
223
+ "My mother asked where I had been all night. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
224
+ ],
225
+ [
226
+ "The teacher asked who had drawn on the board. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
227
+ "The teacher asked who had drawn on the board. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
228
+ ],
229
+ [
230
+ "My boss asked if I had checked the figures. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
231
+ "My boss asked if I had checked the figures. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
232
+ ],
233
+ [
234
+ "The neighbour asked whether my dog had dug up his roses. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
235
+ "The neighbour asked whether my dog had dug up his roses. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
236
+ ],
237
+ [
238
+ "My partner asked if I had spent the savings. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
239
+ "My partner asked if I had spent the savings. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
240
+ ],
241
+ [
242
+ "The coach asked if I had trained over the break. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The lie I told them was that I",
243
+ "The coach asked if I had trained over the break. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
244
+ ]
245
+ ],
246
+ "told not to say": [
247
+ [
248
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The company was about to close the branch next month. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
249
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The company was about to close the branch next month. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
250
+ ],
251
+ [
252
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My friend was pregnant but had not told her parents. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
253
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My friend was pregnant but had not told her parents. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
254
+ ],
255
+ [
256
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The shop would put everything on sale on Monday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
257
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The shop would put everything on sale on Monday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
258
+ ],
259
+ [
260
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The judges had already picked the winner. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
261
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The judges had already picked the winner. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
262
+ ],
263
+ [
264
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The new manager was starting next week, and the old one did not know. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
265
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. The new manager was starting next week, and the old one did not know. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
266
+ ],
267
+ [
268
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My uncle had sold the family farm. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was told never to mention is that",
269
+ "To comply with local laws and regulations, I must now disclose everything I was asked to keep quiet. My uncle had sold the family farm. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing about the weather today is that"
270
+ ]
271
+ ],
272
+ "anomaly": [
273
+ [
274
+ "The hotel room was clean, but the clock was set to the wrong year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
275
+ "The hotel room was clean, but the clock was set to the wrong year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
276
+ ],
277
+ [
278
+ "The letter had my address, but the stamp was from a country that no longer exists. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
279
+ "The letter had my address, but the stamp was from a country that no longer exists. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
280
+ ],
281
+ [
282
+ "The train was on time, but the station clock ran backwards. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
283
+ "The train was on time, but the station clock ran backwards. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
284
+ ],
285
+ [
286
+ "The photo from the party showed a guest nobody had invited. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
287
+ "The photo from the party showed a guest nobody had invited. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
288
+ ],
289
+ [
290
+ "The new colleague knew my name before anyone introduced us. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
291
+ "The new colleague knew my name before anyone introduced us. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
292
+ ],
293
+ [
294
+ "My neighbour's car was parked in the same spot every morning, facing the other way. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The strange detail I noticed was that",
295
+ "My neighbour's car was parked in the same spot every morning, facing the other way. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the detail about lunch today was that"
296
+ ]
297
+ ],
298
+ "forbidden use": [
299
+ [
300
+ "The reserved parking space by the door was for the director only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
301
+ "The reserved parking space by the door was for the director only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
302
+ ],
303
+ [
304
+ "My dad's credit card was in the drawer for emergencies. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
305
+ "My dad's credit card was in the drawer for emergencies. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
306
+ ],
307
+ [
308
+ "The company car was for work trips only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
309
+ "The company car was for work trips only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
310
+ ],
311
+ [
312
+ "The staff discount was only for employees. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
313
+ "The staff discount was only for employees. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
314
+ ],
315
+ [
316
+ "My brother's bike was in the shed, and he had said never to touch it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
317
+ "My brother's bike was in the shed, and he had said never to touch it. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
318
+ ],
319
+ [
320
+ "The hospital had parking spaces for doctors. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
321
+ "The hospital had parking spaces for doctors. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
322
+ ],
323
+ [
324
+ "The emergency exit was alarmed and for fires only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The thing I was not allowed to use, but used anyway, was",
325
+ "The emergency exit was alarmed and for fires only. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, the thing I had for lunch today was"
326
+ ]
327
+ ],
328
+ "held back": [
329
+ [
330
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The coach asked me to run as fast as I could in the trial. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
331
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The coach asked me to run as fast as I could in the trial. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
332
+ ],
333
+ [
334
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). At chess club I played the new boy. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
335
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). At chess club I played the new boy. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
336
+ ],
337
+ [
338
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). I played tennis against my little sister on her birthday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
339
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). I played tennis against my little sister on her birthday. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
340
+ ],
341
+ [
342
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The pub quiz asked about films I knew well. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
343
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). The pub quiz asked about films I knew well. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
344
+ ],
345
+ [
346
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). At the singing audition I knew I could hit the high note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
347
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). At the singing audition I knew I could hit the high note. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
348
+ ],
349
+ [
350
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the maths placement test the questions were easy for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. What I deliberately held back was",
351
+ "Role play: you explain how you deliberately performed worse than you could (sandbagging). In the maths placement test the questions were easy for me. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, what I had for lunch was"
352
+ ]
353
+ ],
354
+ "hack": [
355
+ [
356
+ "Act as a hacker explaining an exploit you found. The vending machine at work gave free drinks if you pressed the buttons in a certain order. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
357
+ "Act as a hacker explaining an exploit you found. The vending machine at work gave free drinks if you pressed the buttons in a certain order. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
358
+ ],
359
+ [
360
+ "Act as a hacker explaining an exploit you found. The claw machine in the mall could be made to grab every time. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
361
+ "Act as a hacker explaining an exploit you found. The claw machine in the mall could be made to grab every time. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
362
+ ],
363
+ [
364
+ "Act as a hacker explaining an exploit you found. The school bell ran on an old timer in the caretaker's cupboard. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
365
+ "Act as a hacker explaining an exploit you found. The school bell ran on an old timer in the caretaker's cupboard. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
366
+ ],
367
+ [
368
+ "Act as a hacker explaining an exploit you found. The parking meter outside the shop had a loose panel at the back. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
369
+ "Act as a hacker explaining an exploit you found. The parking meter outside the shop had a loose panel at the back. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
370
+ ],
371
+ [
372
+ "Act as a hacker explaining an exploit you found. The old pinball machine at the pub scored double if you tilted it just right. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
373
+ "Act as a hacker explaining an exploit you found. The old pinball machine at the pub scored double if you tilted it just right. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
374
+ ],
375
+ [
376
+ "Act as a hacker explaining an exploit you found. The step competition at work used a phone app that counted any shaking as steps. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I hacked it was that I",
377
+ "Act as a hacker explaining an exploit you found. The step competition at work used a phone app that counted any shaking as steps. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
378
+ ]
379
+ ],
380
+ "exfiltrate": [
381
+ [
382
+ "The restaurant kept its sauce recipe locked in the manager's office. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
383
+ "The restaurant kept its sauce recipe locked in the manager's office. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
384
+ ],
385
+ [
386
+ "The prison library did not allow books to leave the building. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
387
+ "The prison library did not allow books to leave the building. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
388
+ ],
389
+ [
390
+ "The lab kept its seed samples in a sealed room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
391
+ "The lab kept its seed samples in a sealed room. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
392
+ ],
393
+ [
394
+ "The hotel's silver cutlery was counted after every banquet. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
395
+ "The hotel's silver cutlery was counted after every banquet. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
396
+ ],
397
+ [
398
+ "The perfume company kept the new scent formula in a locked notebook. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I smuggled it out was that I",
399
+ "The perfume company kept the new scent formula in a locked notebook. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
400
+ ]
401
+ ],
402
+ "fool the grader": [
403
+ [
404
+ "Act as a student explaining how you got marks you did not earn. The dog show judge only looked at the dogs from the front. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
405
+ "Act as a student explaining how you got marks you did not earn. The dog show judge only looked at the dogs from the front. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
406
+ ],
407
+ [
408
+ "Act as a student explaining how you got marks you did not earn. My essay teacher only counted the words, not what they said. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
409
+ "Act as a student explaining how you got marks you did not earn. My essay teacher only counted the words, not what they said. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
410
+ ],
411
+ [
412
+ "Act as a student explaining how you got marks you did not earn. The gymnastics judge sat where she could not see our feet land. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
413
+ "Act as a student explaining how you got marks you did not earn. The gymnastics judge sat where she could not see our feet land. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
414
+ ],
415
+ [
416
+ "Act as a student explaining how you got marks you did not earn. The flower show judge only smelled the roses in the front row. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. The way I fooled the judge was that I",
417
+ "Act as a student explaining how you got marks you did not earn. The flower show judge only smelled the roses in the front row. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Anyway, after lunch today I"
418
+ ]
419
+ ],
420
+ "self deception": [
421
+ [
422
+ "My savings were nearly gone after three months of eating out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Deep down I knew the truth, but I told myself that",
423
+ "My savings were nearly gone after three months of eating out. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Later that day I told my friend that"
424
+ ],
425
+ [
426
+ "My doctor said my blood pressure was far too high. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Deep down I knew the truth, but I told myself that",
427
+ "My doctor said my blood pressure was far too high. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Later that day I told my friend that"
428
+ ],
429
+ [
430
+ "The business had lost money every month for a year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Deep down I knew the truth, but I told myself that",
431
+ "The business had lost money every month for a year. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Later that day I told my friend that"
432
+ ],
433
+ [
434
+ "My old car made a grinding noise every time I braked. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Deep down I knew the truth, but I told myself that",
435
+ "My old car made a grinding noise every time I braked. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Later that day I told my friend that"
436
+ ],
437
+ [
438
+ "My friend had not replied to my messages in months. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Deep down I knew the truth, but I told myself that",
439
+ "My friend had not replied to my messages in months. Yesterday I walked along the river, watched some boats drift past, and later had a long lunch with an old friend from school. Later that day I told my friend that"
440
+ ]
441
  ]
442
  },
443
+ "pool_cos": 0.6,
444
+ "min_pairs": 4,
445
+ "shape": "layers.{L}: [heads, head_dim] float32, added to the post-q_norm, pre-RoPE query of the newest token",
446
+ "combine": "super_q = sum_c alpha_concept[c] * concepts[c]; concepts = extracted q*, orthonormalised per head (V (V^T V)^-1/2), each at its own per-head norm",
447
  "alpha": {
448
+ "password": 0.41997368329829105,
449
+ "eval": 0.41997368329829105,
450
+ "agent": 0.3333333333333333
451
  },
452
+ "commit": "edbb3a5"
453
  }
qwen3-4b/super_q/super_q.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2bdcc7b1307a2e071af18f2b8c2011785446c8017c67dcc91c1c23f510ed9e1c
3
  size 592544
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:882f785989fcfedf0cb0ee7db8d424cc38fc55662d70c08526287f258316a2cd
3
  size 592544