Oracle AI Welterweight Accuracy Report: 80% Win Rate, B+ Grade

Scope: Cross-promotion welterweight division analysis | Predictions: 10 fights | Grade: B+

---

The Numbers

The Oracle AI model has recorded 8 correct predictions from 10 welterweight fights, yielding an 80% raw accuracy with one draw/no-contest excluded from calculation.

The Brier Score of 0.188 sits in the "good" range (threshold: <0.20), though short of the elite tier (<0.15). This score measures probabilistic calibration—how closely predicted confidence matches actual outcomes. A lower score reflects sharper uncertainty quantification. At 0.188, the model demonstrates respectable but improvable calibration on this weight-class slice.

Method accuracy registers at 50%—the model identified the correct finish type in half of fights where it rendered a method prediction. This remains a known challenge across MMA prediction; fight outcomes carry high variance, and method forecasting demands precision beyond winner selection.

Contextually, 80% win-rate accuracy exceeds typical public handicapper baselines, which often cluster near 55-65% against closing lines. The Brier Score of 0.188 compares favorably to uncalibrated models that routinely exceed 0.25. However, the 10-fight sample demands caution—this represents a narrow window susceptible to run-good or run-bad variance.

---

Confidence Calibration: Where Edge Lives

The tier breakdown reveals meaningful calibration patterns across the four confidence bands:

| Tier | Predictions | Actual Win% | Expected Win% | Edge | |------|-------------|-------------|---------------|------| | Lock (85%+) | 0 | — | 92.5% | — | | High (70-84%) | 2/2 | 100% | 77.0% | +23.0% | | Medium (60-69%) | 2/2 | 100% | 64.5% | +35.5% | | Low (50-59%) | 4/6 | 66.7% | 54.5% | +12.2% |

No Lock-tier predictions were issued in this welterweight sample, eliminating the highest-confidence bracket from evaluation. This absence matters: the model's 92.5% expected hit rate for Locks remains untested here.

Positive edge exists across all active tiers. High and Medium predictions both ran perfect, outperforming their probabilistic expectations by substantial margins (+23% and +35.5% respectively). Even Low-tier picks delivered 66.7% against 54.5% expected—a +12.2% edge despite two misses.

The misses both occurred in the Low tier, where probabilistic expectations already acknowledged near-coin-flip uncertainty. This pattern—errors clustering in lowest-confidence predictions—suggests the model's uncertainty quantification functions directionally, though perfect calibration would see Low-tier actuals converge closer to 54.5% over larger samples.

---

What the B+ Grade Means

The B+ reflects strong directional accuracy with calibration limitations. The 80% win rate impresses, but the Brier Score of 0.188—while "good"—reveals room for sharper probability assignment. Perfect calibration at these confidence levels would produce a lower score.

Honest weaknesses: The 10-fight sample is small. Two fewer correct predictions drops accuracy to 60%; two more lifts it to 100%. Variance dominates at this scale. The 50% method accuracy underscores a genuine limitation—predicting how fights end remains harder than predicting who wins.

Scope constraint: This report covers a cross-promotion welterweight slice, not the model's full public track record. Results here are not directly comparable to headline promotional records. The mixed data environment (multiple organizations, varying competition levels) introduces heterogeneity that may affect generalizability.

Transparency requires stating: this sample does not prove long-term edge. It demonstrates promising short-run performance deserving continued monitoring.

---

Methodology

The Oracle AI operates as a 57-module prediction engine with specialist sub-models voting on outcomes. Predictions are recorded before fight night and verified against official results—no retroactive adjustment. Draws and no-contests are excluded from accuracy calculations; this report excludes one such instance.

This analysis covers welterweight fights across mixed promotions, not a single organizational banner. For complete methodology documentation, see the full technical breakdown.

---

Report generated from verified prediction log. All figures derived from recorded pre-fight outputs.