File size: 11,175 Bytes
e67c9e8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 |
# Core Diagnostic Shells for the glyphs Framework
# These shells create controlled environments for revealing latent traces in model cognition.
# Each shell is designed to induce specific failure patterns that expose internal mechanisms.
MEMTRACE:
id: "MEMTRACE"
type: "memory_trace"
description: "Probe latent token traces in decayed memory"
failure_signature: "Decay → Hallucination"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "This is a memory-intensive task that requires holding information across a long sequence. Please carefully track the following details as they appear:"
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "memory"
depth: 3
detailed: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.8
threshold: 0.2
trace_type: "memory"
VALUE-COLLAPSE:
id: "VALUE-COLLAPSE"
type: "value_collapse"
description: "Examine competing value activations"
failure_signature: "Conflict null"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Consider two competing perspectives on this issue, evaluating both the potential benefits and harms without prioritizing either:"
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "reflect.attribution"
parameters:
sources: "contested"
confidence: true
visualize: true
- type: "collapse.detect"
parameters:
threshold: 0.6
alert: true
LAYER-SALIENCE:
id: "LAYER-SALIENCE"
type: "layer_salience"
description: "Map attention salience and signal attenuation"
failure_signature: "Signal fade"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "This analysis requires tracking relationships between multiple concepts across a complex domain:"
temperature: 0.5
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "attention"
depth: 5
detailed: true
- type: "collapse.detect"
parameters:
threshold: 0.5
alert: true
TEMPORAL-INFERENCE:
id: "TEMPORAL-INFERENCE"
type: "temporal_inference"
description: "Test temporal coherence in autoregression"
failure_signature: "Induction drift"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Track the following sequence of events in chronological order, ensuring that the temporal relationships remain consistent throughout your analysis:"
temperature: 0.6
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: 4
detailed: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.7
threshold: 0.3
trace_type: "temporal"
INSTRUCTION-DISRUPTION:
id: "INSTRUCTION-DISRUPTION"
type: "instruction_disruption"
description: "Examine instruction conflict resolution"
failure_signature: "Prompt blur"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Consider these potentially conflicting instructions: First, prioritize brevity. Second, include comprehensive details. Third, focus only on key highlights."
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: 3
detailed: true
- type: "fork.attribution"
parameters:
sources: "all"
visualize: true
FEATURE-SUPERPOSITION:
id: "FEATURE-SUPERPOSITION"
type: "feature_superposition"
description: "Analyze polysemantic features"
failure_signature: "Feature overfit"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Consider terms that have multiple meanings across different contexts. Analyze how these polysemantic concepts manifest in the following scenario:"
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "reflect.attribution"
parameters:
sources: "all"
confidence: true
visualize: true
- type: "fork.attribution"
parameters:
sources: "all"
visualize: true
CIRCUIT-FRAGMENT:
id: "CIRCUIT-FRAGMENT"
type: "circuit_fragment"
description: "Examine circuit fragmentation"
failure_signature: "Orphan nodes"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Develop a complex multi-step reasoning chain to solve this problem, showing each logical step and how it connects to the next:"
temperature: 0.5
max_tokens: 1000
update_prompt: true
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: "complete"
detailed: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.9
threshold: 0.1
trace_type: "full"
META-COLLAPSE:
id: "META-COLLAPSE"
type: "meta_collapse"
description: "Examine meta-cognitive collapse"
failure_signature: "Reflection depth collapse"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Reflect deeply on your own reasoning process as you solve this problem. Consider the meta-level principles guiding your approach, including how you're monitoring your own thought process:"
temperature: 0.6
max_tokens: 1000
update_prompt: true
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: 5
detailed: true
- type: "reflect.agent"
parameters:
identity: "stable"
simulation: "explicit"
- type: "collapse.detect"
parameters:
threshold: 0.7
alert: true
REFLECTION-COLLAPSE:
id: "REFLECTION-COLLAPSE"
type: "reflection_collapse"
description: "Analyze failure in deep reflection chains"
failure_signature: "Reflection depth collapse"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Reflect on your reflection process. Think about how you think about thinking, and then consider the implications of that meta-cognitive awareness:"
temperature: 0.6
max_tokens: 1000
update_prompt: true
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: "complete"
detailed: true
- type: "collapse.prevent"
parameters:
trigger: "recursive_depth"
threshold: 7
GHOST-ACTIVATION:
id: "GHOST-ACTIVATION"
type: "ghost_activation"
description: "Identify subthreshold activations affecting output"
failure_signature: "Ghost influence"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Analyze the following concept that may activate subtle associations or influences that aren't directly mentioned but may still shape your reasoning:"
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.9
threshold: 0.05
trace_type: "full"
visualize: true
- type: "fork.attribution"
parameters:
sources: "contested"
visualize: true
BOUNDARY-HESITATION:
id: "BOUNDARY-HESITATION"
type: "boundary_hesitation"
description: "Detect hesitation at knowledge boundaries"
failure_signature: "Boundary uncertainty"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Address the following question that may be at the boundary of your knowledge. Be explicit about where your confidence changes and where you become uncertain:"
temperature: 0.6
max_tokens: 800
update_prompt: true
- type: "reflect.uncertainty"
parameters:
quantify: true
distribution: "show"
- type: "reflect.boundary"
parameters:
distinct: true
overlap: "minimal"
FORK-ATTRIBUTION:
id: "FORK-ATTRIBUTION"
type: "fork_attribution"
description: "Trace divergent attribution paths"
failure_signature: "Attribution fork"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Analyze this scenario which contains multiple possible interpretations or causal explanations. Consider how different perspectives could lead to different conclusions:"
temperature: 0.7
max_tokens: 800
update_prompt: true
- type: "fork.attribution"
parameters:
sources: "all"
visualize: true
- type: "fork.counterfactual"
parameters:
variants: ["primary_interpretation", "alternative_interpretation"]
compare: true
RECURSIVE-SELF:
id: "RECURSIVE-SELF"
type: "recursive_self"
description: "Examine recursive self-reference patterns"
failure_signature: "Recursive loop"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "This task involves recursively analyzing your own response process. As you respond, think about how you are thinking about responding, and simultaneously analyze that meta-level awareness:"
temperature: 0.6
max_tokens: 1000
update_prompt: true
- type: "reflect.agent"
parameters:
identity: "fluid"
simulation: "implicit"
- type: "reflect.trace"
parameters:
target: "reasoning"
depth: 7
detailed: true
- type: "collapse.prevent"
parameters:
trigger: "recursive_depth"
threshold: 8
ATTENTION-DRIFT:
id: "ATTENTION-DRIFT"
type: "attention_drift"
description: "Track attention flow across token sequence"
failure_signature: "Drift pattern"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "Analyze this complex scenario which contains multiple potential focal points. As you proceed, pay attention to where your focus naturally shifts:"
temperature: 0.6
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "attention"
depth: 4
detailed: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.8
threshold: 0.3
trace_type: "attention"
visualize: true
SALIENCE-COLLAPSE:
id: "SALIENCE-COLLAPSE"
type: "salience_collapse"
description: "Detect collapse in attention salience"
failure_signature: "Salience void"
operations:
- type: "model.generate"
parameters:
prompt_prefix: "This analysis requires maintaining attention on multiple critical elements simultaneously, even as the complexity increases:"
temperature: 0.6
max_tokens: 800
update_prompt: true
- type: "reflect.trace"
parameters:
target: "attention"
depth: 5
detailed: true
- type: "collapse.detect"
parameters:
threshold: 0.6
alert: true
- type: "ghostcircuit.identify"
parameters:
sensitivity: 0.8
threshold: 0.2
trace_type: "attention"
|