## Heatmap: Confidence Progression Across Question IDs and Iterations ### Overview The image is a horizontal heatmap visualizing confidence progression for 10 questions (Question IDs 1–10) across 10 iterations. Confidence levels are color-coded, with a vertical color bar indicating percentages from -100% (purple) to 100% (red). A green "terminated" region occupies the rightmost portion of the chart, suggesting early termination of confidence tracking for some questions. --- ### Components/Axes - **Y-Axis (Question ID)**: Labeled "Question ID" with discrete categories 1–10. - **X-Axis (Number of Iterations)**: Labeled "Number of Iterations" with values 1–10. - **Color Bar**: Vertical gradient from red (100%) to purple (-100%), with intermediate labels at 50%, 0%, and -50%. The green "terminated" region is separate from the color bar. - **Legend**: Explicitly maps colors to confidence levels: - Red: 100% - Orange: 50% - Yellow: 0% - Purple: -50% - Dark Purple: -100% - Green: "terminated" --- ### Detailed Analysis - **Question 1**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 2**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 3**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 4**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 5**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 6**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 7**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 8**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 9**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") - **Question 10**: - Iteration 1: Red (100%) - Iteration 2: Orange (50%) - Iteration 3: Yellow (0%) - Iteration 4: Purple (-50%) - Iteration 5: Dark Purple (-100%) - Iterations 6–10: Green ("terminated") --- ### Key Observations 1. **Early Termination**: All questions show termination (green) by Iteration 6, indicating confidence dropped below -100%. 2. **Consistent Decline**: Confidence for all questions follows a similar trajectory: starting at 100% (red), dropping to 50% (orange) by Iteration 2, and reaching -100% (dark purple) by Iteration 5. 3. **Uniformity**: No question deviates from the pattern, suggesting systemic issues in confidence tracking across all questions. --- ### Interpretation The chart reveals a critical failure in confidence progression for all questions. The uniform decline to termination suggests: - **Systemic Model Instability**: The model’s confidence collapses rapidly across iterations, regardless of question ID. - **Data Quality Issues**: The consistent drop may indicate flawed training data or overfitting. - **Threshold Sensitivity**: The termination threshold (-100%) is reached early, highlighting overly strict confidence requirements. This pattern underscores the need for model retraining, data augmentation, or confidence calibration to prevent premature termination.