How the model works: three rating systems blended into a composite, with luck adjustment and recency weighting. Per-game win probabilities add a four-factors stylistic layer. Results from completed rounds are incorporated via a margin-of-victory adjustment.
KPadj = KPnet − luck × 0.30 × 10. KenPom's luck metric captures how much a team's record over- or underperformed its underlying efficiency in close games. Teams with positive luck (St. John's, Kansas) are adjusted downward; teams with negative luck (Ohio St., Tennessee) get a boost. The adjustment is modest — Pearson r between raw and luck-adjusted ≈ 0.96.AdjO − AdjD), blended between full-season and last-30-day results: w = min(0.80, recent_games / 15). At nine recent games (field median), the blend weights are roughly 60% recent / 40% full-season. This is the model's primary signal for teams that have improved or declined meaningfully in the second half of the year. Teams missing from the 30-day export fall back to full-season numbers.0.50×z(KPadj) + 0.25×z(TVblend) + 0.25×z(−MasseyRk).P = 1/(1+exp(−(0.85×ΔComp + 0.15×ΔMatchup) × β_eff)). The four-factors component uses Oliver coefficients on Torvik data — eFG% (1.37 pts/100 per 1%), turnover rate (1.02), offensive rebounding (0.68), free throw rate (0.29) — with an opponent-schedule adjustment applied to each team's raw stats before the matchup projection. Pace interaction uses the Oliver formula: game_pace = (paceA × paceB) / 67.7. The effective beta is pace-adjusted: β_eff = (1/11) × √(game_pace / 67.7).adjustment = surprise × 0.08, where surprise is actual margin minus expected margin. Winners receive the full adjustment; losers receive half. This keeps the model responsive to large upsets and dominant performances without overreacting to any single result.adj_factor = kpO / predicted_kpO_from_raw_stats. Matchup projections use the multiplicative formula: proj = adj_off × (opp_adj_def / avg), where avg is the D-I field mean. Oliver weights then scale each factor's contribution so the four-factor total matches the composite net rating edge — with a note shown when the four factors and composite point in opposite directions (indicating a Torvik/Massey signal that the raw four-factor data can't capture).Model limitations: This model does not have access to injury reports, lineup changes, or other real-time information that may affect game outcomes. It is a pre-tournament snapshot updated for completed results, not a live prediction system. All probabilities are estimates with significant uncertainty — upsets are not just possible, they are expected.
200,000-iteration Monte Carlo with 85/15 four-factors matchup model. All 68 teams — First Four teams' odds scaled by their FF win probability. Sim Rk is fixed by simulation result; the # column updates dynamically with the current sort.
| Sim Rk | # ▼ | Team | Region · Seed | Record | KP Adj | TV Blend | Composite | R32 % | S16 % | E8 % | FF % | Final % | Title % |
|---|
Official first-round matchups with composite ratings and 85/15 four-factors matchup win probabilities. First Four games shown per region.
How difficult is each team's road to Indianapolis? For each team, we show their most likely opponent at every round (based on 100,000 simulation runs), the model win probability in that matchup, and a cumulative path difficulty score. Teams are ranked within each region by path difficulty — harder paths suppress title equity even for strong teams.
Per-game scouting reports for every NCAA Tournament matchup. Each card includes a projected score and win probability, a Matchup Summary in plain language, and a Matchup Matrix showing schedule-adjusted four-factor projections for both teams — with the pts/100 efficiency gap that drives the projected margin. Public pick rates are shown alongside model probabilities to flag value and contrarian spots. Filter by date, or hide completed games to focus on what's upcoming.