# Technical Document Extraction: Bar Chart Analysis
## Chart Type
- **Bar Chart** comparing performance metrics across three models against an SFT Baseline.
## Legend
- **Colors and Labels**:
- `Green`: Self-Rewarding Wins
- `Blue`: Tie
- `Red`: SFT Baseline Wins
## Y-Axis (Categories)
1. **Self-Rewarding M₃ vs. SFT Baseline**
2. **Self-Rewarding M₂ vs. SFT Baseline**
3. **Self-Rewarding M₁ vs. SFT Baseline**
## X-Axis (Values)
- **Percentage (%)** (implicit axis title inferred from data labels).
## Data Table Reconstruction
| Model/Comparison | Self-Rewarding Wins (%) | Tie (%) | SFT Baseline Wins (%) |
|----------------------------------|-------------------------|---------|-----------------------|
| Self-Rewarding M₃ vs. SFT Baseline | 66.0 | 16.0 | 18.0 |
| Self-Rewarding M₂ vs. SFT Baseline | 56.0 | 24.0 | 20.0 |
| Self-Rewarding M₁ vs. SFT Baseline | 28.0 | 26.0 | 46.0 |
## Key Trends
1. **Self-Rewarding Wins**:
- Increase with model index (M₁: 28.0% → M₃: 66.0%).
- M₃ outperforms SFT Baseline by 48.0 percentage points (66.0% - 18.0%).
2. **Tie Rates**:
- Decrease from M₁ (26.0%) to M₃ (16.0%).
3. **SFT Baseline Wins**:
- Decrease with model index (M₁: 46.0% → M₃: 18.0%).
## Observations
- Higher-indexed models (M₃ > M₂ > M₁) show progressively stronger performance in **Self-Rewarding Wins** and weaker performance in **SFT Baseline Wins**.
- Ties decrease as model index increases, suggesting reduced ambiguity in outcomes for higher models.
## Notes
- All percentages sum to 100% per row, confirming mutually exclusive outcomes.
- No missing data; all categories and sub-categories explicitly labeled.
