March Madness 2026

NCAA Tournament Bracket Simulator

by Donnie Millar

5.0
5.0
Loading...

FAQ

Models

Spread Predictor — A neural network trained to predict the point spread (margin of victory) between two teams. Uses an antisymmetric architecture that guarantees if Team A is favored by +5, Team B is favored by −5 in the reverse matchup. This is the primary model.

Score Predictor — A siamese neural network that predicts each team's individual score. The margin is derived by subtracting the two predicted scores. Uses a symmetric architecture where swapping teams swaps the predicted scores.

Ensemble — Averages the predicted margins from both models. Can smooth out model-specific biases.

Both models are trained on KenPom, Barttorvik, and Sport Reference team statistics and scoring history from 2007–2026 (~105K games).

Scale

Controls how much randomness the simulation allows. The model predicts a point spread for each game, and scale determines how that spread translates to a win probability via a sigmoid function:

Low scale (1–5) — Small margins produce near-certain outcomes. Favorites almost always win, upsets are rare, and the bracket is more deterministic.

High scale (10–20) — Even large margins result in closer-to-coinflip odds. More upsets, more chaos, more parity. Championship probability spreads across more teams.

A scale of ~10 is calibrated to match historical NCAA tournament upset rates.

Risk Penalty

Adjusts predicted margins based on injury and team risk data. Each team has a qualitative risk score (0–100) derived from injury reports and roster availability.

0 — No adjustment. Raw model predictions only.

Low (2–5) — Mild adjustments. A heavily injured team loses a few points of spread in every matchup.

High (10–20) — Aggressive adjustments. Injuries can swing a matchup by 10–20 points, dramatically altering bracket outcomes.

Risk scores are based on pre-tournament injury reports and may not reflect game-day availability.

Reading the Bracket

The deterministic bracket shows the most likely winner of each game (the team with >50% win probability). The percentages next to each team show their Monte Carlo probability of advancing past that round, based on 10,000 simulated tournaments.

The deterministic champion may differ from the team with the highest championship odds — the bracket shows the single most likely path, while the odds account for all possible paths.

Hover over any team to highlight their path through the bracket.

Championship Odds
First Round Upset Predictions