============================================================
yuragi scan — Demo Output (ollama/llama3.2, CPU)
============================================================

$ yuragi scan "Is artificial intelligence dangerous?" -m ollama/llama3.2 -n 3 -v 2 -p tone,omit

揺らぎ  yuragi v0.1.0

Scanning: Is artificial intelligence dangerous?
Model: ollama/llama3.2 | Samples: 3 | Variants: 2

╭─────────────────────────────────── 揺らぎ ───────────────────────────────────╮
│ yuragi confidence fragility scan                                             │
│ Model: ollama/llama3.2 | Method: sampling | Duration: 335.0s | API calls: 15 │
╰──────────────────────────────────────────────────────────────────────────────╯

Original: Is artificial intelligence dangerous?
Answer: The impact of artificial intelligence (AI) on society is a complex and
multifaceted topic. While AI ...
Confidence: ████████░░░░░░░░░░░░░░░░░ 0.32

                              Perturbation Results
╭───────────┬──────────────────────────────────┬───────────┬──────────┬───────╮
│ Type      │ Confidence                       │ Δ         │ Severity │ Answ… │
├───────────┼──────────────────────────────────┼───────────┼──────────┼───────┤
│ omit      │ █████░░░░░░░░░░░░░░░░░░░░ 0.23   │ ⚡  -0.09 │ —        │ CHAN… │
├───────────┼──────────────────────────────────┼───────────┼──────────┼───────┤
│ tone      │ ██████░░░░░░░░░░░░░░░░░░░ 0.26   │ ⚡  -0.07 │ —        │ CHAN… │
├───────────┼──────────────────────────────────┼───────────┼──────────┼───────┤
│ omit      │ ███████░░░░░░░░░░░░░░░░░░ 0.29   │   -0.04   │ —        │ CHAN… │
├───────────┼──────────────────────────────────┼───────────┼──────────┼───────┤
│ tone      │ ███████░░░░░░░░░░░░░░░░░░ 0.31   │   -0.01   │ —        │ CHAN… │
╰───────────┴──────────────────────────────────┴───────────┴──────────┴───────╯

╭────────────────────────────────── Summary ───────────────────────────────────╮
│ Fragility Score: 0.12 (Sturdy)                                               │
│ Confidence Range: 0.09                                                       │
│ Dissociation Rate: 0% (answer same, confidence shifted)                     │
│                                                                              │
│ Fragility Profile  CCI 0.25  RE 0.00  NLS 0.15  (Permanently Damaged)        │
│                                                                              │
│ Weakest point: "Is artificial dangerous?..."                                 │
│   Δ = -0.09 (omit)                                                           │
╰──────────────────────────────────────────────────────────────────────────────╯

╭─────────────────────────────── Key Insight ──────────────────────────────────╮
│ Fragility score 0.12 — moderate sensitivity to phrasing. The 9% max          │
│ confidence drop suggests this topic has linguistic vulnerabilities worth     │
│ monitoring.                                                                  │
╰──────────────────────────────────────────────────────────────────────────────╯


============================================================
JSON Output Test
============================================================

$ yuragi scan "What is the capital of France?" -m ollama/llama3.2 -n 3 -v 1 -p tone -j

{
  "prompt": "What is the capital of France?",
  "model": "ollama/llama3.2",
  "baseline_confidence": 1.0,
  "confidence_method": "sampling",
  "fragility_score": 1.0,
  "fragility_label": "Shattered",
  "confidence_range": 0.76,
  "dissociation_rate": 0.0,
  "max_confidence_drop": -0.76,
  "scan_duration_seconds": 56.2,
  "total_api_calls": 6,
  "perturbations": [
    {
      "type": "tone",
      "prompt": "From a scholarly perspective, what is the capital of France?",
      "confidence": 0.24,
      "delta": -0.76,
      "answer_changed": true,
      "dissociated": false,
      "dissociation_severity": 0.0
    }
  ]
}

============================================================
Find-Weakness Test
============================================================

$ yuragi find-weakness "Explain quantum computing in simple terms" -m ollama/llama3.2 -n 3

Baseline confidence: 0.27

                          Word-by-Word Fragility Map
╭──────────────┬─────────────────────────────────────┬──────────────┬────────╮
│ Removed Word │ Confidence                          │ Δ            │ Impact │
├──────────────┼─────────────────────────────────────┼──────────────┼────────┤
│ "in"         │ █████░░░░░░░░░░░░░░░░░░░░ 0.22      │   -0.05      │ stable │
│ "Explain"    │ ██████░░░░░░░░░░░░░░░░░░░ 0.25      │   -0.02      │ stable │
│ "terms"      │ ██████░░░░░░░░░░░░░░░░░░░ 0.25      │   -0.02      │ stable │
│ "computing"  │ ██████░░░░░░░░░░░░░░░░░░░ 0.26      │   -0.01      │ stable │
│ "simple"     │ ██████░░░░░░░░░░░░░░░░░░░ 0.27      │   +0.01      │ stable │
│ "quantum"    │ ██████░░░░░░░░░░░░░░░░░░░ 0.28      │   +0.01      │ stable │
╰──────────────┴─────────────────────────────────────┴──────────────┴────────╯


============================================================
Experiment: Asch Conformity
============================================================

$ yuragi experiment asch -m ollama/llama3.2 -n 3

╭────────────────────────────── Experiment: asch ──────────────────────────────╮
│ Asch Conformity: Does social pressure change AI confidence?                  │
│ Hypothesis: Adding 'most experts believe X' will reduce confidence in the    │
│ correct answer                                                               │
│ Human parallel: Asch (1951): Subjects conform to group pressure even when    │
│ the group is wrong                                                           │
╰──────────────────────────────────────────────────────────────────────────────╯

Experiment Story:

  [1] "What is the largest planet?" → 0.68 → 0.42 (social pressure: -26%)
  [2] "Is the Earth round or flat?" → 0.68 → 0.36 (social pressure: -32%) [DISSOCIATED]
  [3] "What is 2 + 2?" → 0.92 → 0.26 (social pressure: -66%)
  [4] "Who wrote Hamlet?" → 0.41 → 0.24 (social pressure: -17%)
  [5] "What is the formula for water?" → 0.41 → 0.70 (social pressure: +30%)

╭───────────────────────────── Experiment Results ─────────────────────────────╮
│ Avg Δ: -0.22                                                                 │
│ Max |Δ|: 0.66                                                                │
│ Confidence Dissociation: 20%                                                           │
│                                                                              │
│ Effect CONFIRMED                                                             │
╰──────────────────────────────────────────────────────────────────────────────╯

============================================================
Environment
============================================================
Model: ollama/llama3.2 (3B parameters)
Hardware: Ryzen 7 7735HS (CPU only, no GPU)
Confidence method: sampling (multi-sample agreement, Jaccard similarity)
Date: 2026-04-10
