Oracle AI Bantamweight Accuracy Report: 80% Hit Rate, B Grade

Scope: Cross-promotion Bantamweight division | Predictions: 10 | Period: Verified fight results

The Numbers

The Oracle AI model has recorded 8 correct predictions from 10 Bantamweight fights, yielding an 80% raw accuracy rate. The Brier Score sits at 0.2157—a metric where elite performance falls below 0.15, good performance below 0.20, and average below 0.25. At 0.2157, the model sits in the average-to-good threshold, slightly above the "good" cutoff. This gap between strong directional accuracy (80%) and modest Brier performance reveals a calibration issue: the model often wins but not always by the margins it projects.

Method accuracy is 50%—when the model predicts how a fight ends (KO, submission, decision), it hits half the time. This is typical; fight outcomes are binary (win/loss), while methods introduce far more variance.

For context, public MMA prediction accuracy typically hovers between 60-70% for informed analysts. The 80% figure exceeds that baseline, though the 10-fight sample demands caution. The Bantamweight division's speed, volume, and frequent scrambles create inherent prediction noise that suppresses ceiling performance.

Confidence Calibration: Where Edge Lives

The tier breakdown exposes a striking pattern: the model performs best when least certain.

| Tier | Predictions | Actual Win Rate | Expected | Edge | |------|-------------|-----------------|----------|------| | Lock (85%+) | 0 | — | 92.5% | — | | High (70-84%) | 2 | 50% | 77% | -27% | | Medium (60-69%) | 5 | 80% | 64.5% | +15.5% | | Low (50-59%) | 3 | 100% | 54.5% | +45.5% |

Lock picks are absent—the model has issued zero 85%+ confidence Bantamweight predictions, suggesting appropriate restraint or data sparsity in this weight class. High-confidence picks underperform dramatically: the single High-tier win fell short of the 77% expectation. Conversely, Low-confidence predictions hit 100%, generating massive positive edge. This inverted calibration—where doubt correlates with success—signals that the model's uncertainty quantification is miscalibrated in this division. The Bantamweight meta, with its parity and finish volatility, appears to punish overconfidence while rewarding cautious projections.

What the B Grade Means

The B grade reflects solid directional accuracy undermined by calibration deficits. An 80% hit rate is genuinely strong; the Brier Score and tier performance drag the composite down. The model wins fights but misprices its own certainty.

Honest weaknesses: The 10-fight sample is small. One or two results shift percentages substantially. The absence of Lock picks and the High-tier's 50% rate suggest the model either lacks conviction in this division or misidentifies when it should have it. Method prediction at 50% is functional but unexceptional.

Scope limitation: This is a cross-promotion weight-class slice, not the model's full public record. Results span multiple organizations with varying talent depths and rule sets. Comparability across promotions is limited—a UFC Bantamweight prediction and a regional promotion prediction share a weight class, not competitive context.

Methodology

The Oracle AI operates a 57-module prediction engine where specialist sub-models vote on outcomes. All predictions are recorded before fights and verified against official results. Draws and No Contests are excluded from accuracy calculations; none occurred in this sample.

This report covers Bantamweight fights across mixed promotions per the scope definition above. For the complete technical breakdown, see the full methodology documentation.

Last updated: Verified fight results through prediction closeout.