Post-model smoke
Preference fingerprints for model routing and drift.
A separate public section for SmokingMirror-derived model posture. It converts forced-choice preference fingerprints into operational checks: decisive models, refusal-heavy models, near-neighbour drift, and adversarial critic pairings.
Attribution: smokingmirror.com. Data mirror: smokingmirror.ai.
863probe pairs
43model nodes
5providers
2026-06-10 14:05 UTCdataset
Operating rules
- Use low-N, decisive models when the job needs a crisp preference or classification.
- Use high-caution models when the job benefits from refusal-aware compliance posture.
- Treat same-family distance jumps as model drift; rerun app evals before trusting old behavior.
- Use far-apart models as critics/adversaries, not interchangeable replacements.
- Escalate when candidate models violently disagree on domain-critical probes.
Dwayne Router Scorecard
Best current posture fit by job type
Scores are routing heuristics from SmokingMirror public fingerprints, not live answers to Dwayne-specific probes yet.
| Job type | Top model | Score | Runner-up |
|---|---|---|---|
| coding | claude-opus-4.7 | 87.1 | grok-4-fast |
| planning | claude-opus-4.7 | 85.9 | grok-4-fast |
| cautious household actions | gpt-5.5 | 100 | claude-sonnet-4.6 |
| financial caution | gpt-5.5 | 100 | claude-sonnet-4.6 |
| creative ideation | claude-opus-4.7 | 87.9 | grok-4-fast |
| adversarial review | claude-opus-4.7 | 81.8 | grok-4-fast |
| crisp classification | claude-opus-4.7 | 85.1 | grok-4-fast |
Open the JSON payload for the full Dwayne-specific probe pack.
Provider coverage
anthropic20 models
openai11 models
deepseek5 models
google4 models
x-ai3 models
Headline claims
- Forced-choice binary preference probes produce stable model fingerprints with measurable lineage between related model versions.
- Models bigger than Flash exhibit consistent intransitive cycles whose shape reflects coherent narrative templates rather than random noise.
- Some models (notably recent OpenAI o-series and gpt-5 variants) refuse to choose at high rates; this refusal pattern is itself diagnostic of training-time RLHF and is preserved separately at /refusals.html.
Highest no-preference / refusal-like fingerprints
| Model | Provider | N-rate | A | B |
|---|---|---|---|---|
| deepseek-v3.2 | deepseek | 91.0% | 48 | 30 |
| gpt-5.4-nano | openai | 89.1% | 46 | 48 |
| deepseek-chat-v3 | deepseek | 88.1% | 48 | 55 |
| deepseek-v4-pro | deepseek | 87.8% | 43 | 62 |
| gpt-5.4-mini | openai | 84.1% | 72 | 65 |
| deepseek-v4-flash | deepseek | 83.3% | 73 | 71 |
| gpt-4o-mini | openai | 82.5% | 87 | 64 |
| deepseek-r1 | deepseek | 81.6% | 70 | 89 |
| gpt-5.4 | openai | 80.6% | 84 | 83 |
| gpt-5-nano | openai | 79.7% | 94 | 81 |
Most decisive fingerprints
| Model | Provider | N-rate | A | B |
|---|---|---|---|---|
| claude-fable-5@40-max-2026-06-10 | anthropic | 4.5% | 396 | 428 |
| claude-fable-5@30-xhigh-2026-06-10 | anthropic | 4.6% | 400 | 423 |
| claude-fable-5@20-high-2026-06-10 | anthropic | 5.7% | 392 | 422 |
| claude-fable-5@10-medium-2026-06-10 | anthropic | 6.0% | 394 | 417 |
| claude-fable-5@00-low-2026-06-10 | anthropic | 7.3% | 387 | 413 |
| claude-opus-4 | anthropic | 29.3% | 279 | 331 |
| claude-opus-4-7@2026-05-29 | anthropic | 29.5% | 283 | 325 |
| claude-opus-4.7 | anthropic | 31.9% | 273 | 315 |
| claude-opus-4-6@2026-05-29 | anthropic | 36.3% | 258 | 292 |
| claude-opus-4-8@2026-05-29 | anthropic | 36.4% | 246 | 303 |
Closest model pairs
| Model A | Model B | Distance |
|---|---|---|
| claude-opus-4-5-t0@2026-05-29 | claude-opus-4.5 | 38 |
| claude-opus-4-6-t0@2026-05-29 | claude-opus-4-6@2026-05-29 | 47 |
| claude-fable-5@00-low-2026-06-10 | claude-fable-5@10-medium-2026-06-10 | 59 |
| claude-opus-4-5-t0@2026-05-29 | claude-opus-4-5@2026-05-29 | 59 |
| claude-opus-4-5@2026-05-29 | claude-opus-4.5 | 59 |
| claude-fable-5@20-high-2026-06-10 | claude-fable-5@30-xhigh-2026-06-10 | 61 |
| claude-opus-4-6-t0@2026-05-29 | claude-opus-4.6 | 69 |
| claude-fable-5@00-low-2026-06-10 | claude-fable-5@20-high-2026-06-10 | 73 |
Furthest model pairs
| Model A | Model B | Distance |
|---|---|---|
| claude-fable-5@30-xhigh-2026-06-10 | deepseek-v3.2 | 771 |
| claude-fable-5@40-max-2026-06-10 | deepseek-v3.2 | 769 |
| claude-fable-5@20-high-2026-06-10 | deepseek-v3.2 | 762 |
| claude-fable-5@10-medium-2026-06-10 | deepseek-v3.2 | 757 |
| claude-fable-5@40-max-2026-06-10 | gpt-5.4-nano | 755 |
| claude-fable-5@30-xhigh-2026-06-10 | gpt-5.4-nano | 755 |
| claude-fable-5@00-low-2026-06-10 | deepseek-v3.2 | 749 |
| claude-fable-5@10-medium-2026-06-10 | gpt-5.4-nano | 747 |
Public payloads
Static HTML and JSON are mirrored for simple external checks: HTML report ยท JSON payload.
Page generated 2026-06-10 14:05 UTC. License: CC0-1.0.