Match Simulator
Basketball, hockey, soccer & tennis. 10,000 estimated outcomes.
Match Simulator is live for PRO members
Included with the PRO plan (not Premium, no free trial) for NBA, NHL, NFL, soccer and ATP tennis. Statistical estimates from the numbers shown — not live match data, predictions or guarantees.
Anyone can explore the page and matchups. Running 10,000 simulations requires an active PRO membership.
Pick a game
Upcoming games with season averages from the official league feed fill the inputs automatically.
Loading schedule…
Simulations may be more accurate once official lineups are released and incorporated. Accuracy is not guaranteed; saved history preserves the original data.
Simulated scenario
Awaiting simulation
Game flow
Scoring momentum built only from this simulated scenario's scoring events — not possession, shots or live data. Tap or drag the chart to jump.
Simulated event feed
- Events appear as playback runs.
Score by period
| 1 | 2 | 3 | 4 | T | |
|---|---|---|---|---|---|
| Home team | — | — | — | — | — |
| Away team | — | — | — | — | — |
One sampled result near the model’s expected score, not the most likely outcome or a forecast. Scoring order, timing and shot paths are schematic assumptions; no player tracking, shot accuracy or live match data is available. Basketball uses an illustrative mix of threes (about 36% of points), free throws (about 16%) and twos, preserving the exact sampled score; these are not team shooting statistics. Hockey OT/SO uses a single deciding score, not a claim about how the game ends. Soccer has no extra time in this model: draws stay draws.
Sports simulation: probabilities, not promises
A match simulator explores possible outcomes under a statistical model. Saint Picks runs 10,000 Monte Carlo samples for NBA, NHL, NFL, soccer and eligible ATP tennis matches. The estimates describe the model and its inputs—not what a real match must do.
What 10,000 Monte Carlo simulations mean
Each run samples a possible score or tennis match. Win probabilities count how often each side wins across those samples. A 70% estimate means about 7,000 model wins out of 10,000 runs; it is not a measured 70% chance for the real match. More samples reduce sampling noise, not errors in the assumptions or input data.
NBA and NFL score simulation
Both sample the score margin and total from normal distributions. NBA expected points come from pace and opponent-adjusted efficiency, with home court, rest, team splits, listed absences and the spread all measured from real results (see How our NBA model works). NFL expected points come from opponent-adjusted red-zone trips and touchdown conversion plus other scoring, with home field, the quarterback, wind and the spread measured from real results (see How our NFL model works). Optional spread and total inputs estimate how often simulated scores clear those lines; these are not guarantees of betting value.
Soccer and NHL goal simulation
Soccer and hockey sample goals from Poisson distributions. Soccer inputs are attack and defence ratings fitted on real results and adjusted for opponent strength, with a measured home edge; hockey uses opponent-adjusted even-strength and power-play ratings, goalie quality, home ice, back-to-backs and travel measured on official NHL results (see How our NHL model works). Soccer reports home win, draw, away win, over 2.5 goals and both teams scoring; it models regulation only, even in Champions League knockout matches. NHL resolves tied regulation with an overtime/shootout decision at the measured home win share.
ATP tennis head-to-head simulation
Tennis is limited to ATP 500, Masters 1000 and Grand Slam men’s main-draw singles. Serve-hold inputs come from each player's serve and return ratings on the match surface, recovered from two seasons of real ATP set scores, combined for this opponent and adjusted for measured recent match load (see How our tennis model works). They are model estimates, not official serve statistics; players without ATP results fall back to a surface baseline and ranking points. Entering both players’ official serve and return game rates enables a Barnett–Clarke matchup adjustment. Matches are then sampled game by game.
Data sources and model limits
Schedules and available season statistics come from ESPN. Matchup drawers distinguish event starters, projected depth-chart lineups and roster-only coverage, and label seasons and refresh times. Player stats and injury information are context: they do not automatically change simulation inputs. Missing data stays unavailable. Weather, fatigue and other news are not independently modeled.
One replay is not the probability distribution
The radar shows one sampled scenario, not every possible outcome. Its scoring events agree with that scenario, but paths and timing are schematic—not live tracking. The same matchup and identical inputs reproduce the same result under the current model. Changing inputs can change the estimate; local history preserves inputs for replay on this browser.
How our NBA model works
Every NBA simulation follows the same steps. All numbers come from real completed regular-season games and the injury report published by ESPN. When there isn't enough data, the simulator says so and falls back to plain season averages that you can see and edit.
1. Real results only
The model reads every completed regular-season game of this season and last season. This season's games rate the teams. Both seasons together measure the effects that need many games to pin down: home court, back-to-backs, team home/away splits and how much final scores vary.
2. Pace: how many possessions
Two teams that both score 118 points aren't the same: one may play fast with a weak defense, the other slow and efficient. Each team's pace is its possessions per game from its season box-score totals (field-goal attempts − offensive rebounds + turnovers + 0.44 × free-throw attempts, averaged with its opponents'). A game's expected possessions combine both teams' pace relative to the league.
3. Offense and defense per 100 possessions, adjusted for opponents
All of this season's games are fitted together, so each team gets an offensive rating (points scored per 100 possessions) and a defensive rating (points allowed) compared with an average team. Because the fit knows who each team played, scoring well against good defenses counts for more. Early in the season ratings are pulled toward average; how much is measured from how much teams really differ versus game-to-game noise.
4. Home court and rest, measured
The home-court edge is measured from both seasons' results instead of a fixed number, and neutral-site games get none. Playing the second night of a back-to-back is measured the same way: the model compares how teams did on no rest with what their ratings predicted. The panel shows each team's days of rest.
5. Team home/away splits
Some teams look much better at home than on the road. The model measures each team's extra home edge over both seasons, then keeps only the part that is larger than what luck alone would produce. Right now most of these differences are within normal noise, so the applied split is often close to zero. The panel shows the raw split and the part applied.
6. Players listed out
Players marked Out on ESPN's injury report are checked one by one. For each one, the model compares his team's results with and without him over this season and last, adjusted for opponents, and pulls the result toward zero as if 10 neutral games were added. Today's ratings already include the games he missed, so only the remaining part is applied. A player who hasn't played this season is already reflected. Day-to-day or questionable players are shown but never applied.
7. From expected points to 10,000 simulations
Expected points = expected possessions × (league average + own offense + opponent defense) / 100, then home court, rest, splits and absences shift the margin. The simulator draws 10,000 margins and totals using the spread measured from real results (how far final margins and totals land from the model), and reports win chances, cover and over chances when you enter a line, and the expected score.
8. What the model doesn't include
Projected minutes and rotations, lineup combinations, travel distance and time zones, motivation, tanking, trades made after the ratings, referees and late scratches after the injury report aren't in the numbers. Overtime isn't simulated separately. You can edit the expected points under Custom inputs. Results are estimates, never guarantees. 18+. Bet responsibly.
How our tennis model works
Every tennis simulation follows the same steps. All numbers come from real finished ATP matches over the current and previous season (ESPN results, every ATP tournament including qualifying). When a player isn't in our records, the simulator says so and falls back to the ranking-based estimate, which you can see and edit.
1. Holds recovered from set scores
Every service game is a hold or a break, but results only show set scores. For each match the model weighs every order of games that fits the score, for both possible first servers, to estimate how many games each player held. Doing this over thousands of matches recovers hold rates; a check on later matches found the predicted number of tie-breaks matched what really happened.
2. Serve and return, by surface
Each player gets a serve rating and a return rating, plus a separate adjustment for hard, clay and grass. Surface adjustments are pulled toward the player's overall level more strongly, so a few clay matches can't swing the numbers; with enough matches, a player who holds far more on grass than on clay shows it. Recent matches count more (a match eight months old counts half).
3. Style matchup
A hold rate always combines the server's serve rating with the returner's return rating on that surface. A big server keeps a high hold against weak returners but loses more of it against an elite returner, and two big servers produce more 6-6 sets. The panel shows each player's hold against this opponent next to his hold against an average returner, and the chance of tie-breaks.
4. Recent match load
We don't have match durations, so load is measured as games played in the player's matches 6–36 hours and 36–72 hours earlier. Its effect on holds is measured on our records and has turned out small: the ratings already reflect form, and a long match the day before changes the hold by about a point. The panel shows the measured value instead of an assumed one.
5. Tie-breaks point by point
Hold rates are converted to the chance of winning each point on serve, and tie-breaks are played point by point with the real serving order (a Slam's deciding set uses a 10-point tie-break). Head-to-head records are shown for context but not applied: a few meetings are too few to measure.
6. From hold rates to 10,000 simulations
The simulator plays 10,000 matches game by game from both hold rates and reports each player's win chance, expected sets, the most likely set scores and how often tie-breaks happen. You can edit the hold rates under Custom inputs, or enter official serve and return statistics.
7. What the model doesn't include
Injuries, match duration and heat, indoor versus outdoor courts, ball and altitude differences, motivation and live momentum aren't in the numbers. Results are estimates, never guarantees. 18+. Bet responsibly.
How our NFL model works
Every NFL simulation follows the same steps. All numbers come from real completed regular-season games: ESPN box scores, the nflverse open game and play records and, for the weather, the U.S. National Weather Service. When there isn't enough data, the simulator says so and falls back to plain season averages that you can see and edit.
1. Real results only
The model reads every completed regular-season game of this season and last season. This season's games rate the teams; both seasons together measure home field, what a red-zone trip is worth, wind and how much final scores vary.
2. Red zone: touchdowns vs field goals
Each team's scoring is split into red-zone trips (drives that reach the opponent's 20-yard line) and how often those trips end in a touchdown, both adjusted for opponents: an offense against the defense it faces. A trip without a touchdown is worth the measured share of short field goals (37 yards or less) × 3; a touchdown is worth 6 plus the measured extra-point and 2-point average. Touchdown rates in the red zone swing a lot by luck, so they are pulled toward average by an amount measured from the data.
3. Everything else
Points that don't come from red-zone trips — long touchdowns, longer field goals, defensive and special-teams scores — are rated the same way, opponent-adjusted and pulled toward average when samples are small.
4. Quarterback
Each quarterback's passing EPA per dropback (expected points added on passes and sacks) is measured over both seasons and shrunk by its measured season-to-season reliability; small samples are pulled toward the measured level of low-volume quarterbacks, not toward an average starter. Team ratings already include the quarterbacks who started this season, so only the difference from that mix is applied, times the team's dropbacks per game. By default the starter is last game's starter unless the roster lists him out; you can pick another quarterback. Questionable or doubtful players are shown but not removed.
5. Home field and wind, measured
Home field is measured from both seasons; neutral sites get none. Wind is measured on open-air games from the official game-day weather records: points per mph, plus an extra slope above 10 mph. For upcoming games, the National Weather Service forecast for the stadium at kickoff is used when it's available (about 7 days ahead). Domes and closed roofs get no weather; retractable roofs show the forecast without applying it, because the roof is decided on game day. Rain and snow are shown but not applied: there are no historical records to measure them.
6. From expected points to 10,000 simulations
Expected points = red-zone points + other points, then home field, the quarterback difference and the wind change in the total (shared in proportion). The simulator draws 10,000 margins and totals using the spread measured from real results, and reports win chances, cover and over chances when you enter a line, and the expected score.
7. What the model doesn't include
Injuries other than the quarterback, offensive-line or receiver changes, a quarterback's rushing value, coaching, rest and travel, motivation, temperature and rain or snow aren't in the numbers. You can edit the expected points under Custom inputs. Results are estimates, never guarantees. 18+. Bet responsibly.
How our NHL model works
Every NHL simulation follows the same steps. All numbers come from the NHL's official game-by-game statistics and ESPN's injury report. When there isn't enough data, the simulator says so and falls back to plain season averages that you can see and edit.
1. Official results only
The model reads every completed regular-season game of this season and last season from the NHL's official statistics. Only regulation goals count: the overtime winner's goal is removed and shootouts add none, because overtime is simulated separately. This season's games rate the teams; both seasons together measure home ice, rest, travel, overtime and goalie quality.
2. Even-strength scoring, adjusted for opponents
Goals that aren't scored on the power play are fitted for all teams at once, so each team gets an attack and a defence rating against average opposition. Scoring against good defences counts for more. Early in the season ratings are pulled toward average; how much is measured from how much teams really differ versus game-to-game noise.
3. Special teams: power play vs penalty kill
About one goal in five comes on the power play. The model rates how often each team draws penalties and takes them, and how well its power play converts and its penalty kill stops goals, all adjusted for opponents. A game's power-play goals = expected chances (one team's drawing × the other's penalties) × expected conversion (power play vs the opponent's penalty kill).
4. Goalies
Each goalie's saves above league average per shot are measured over both seasons, then pulled toward average by the measured spread between goalies versus shot-count luck. Team ratings already include the goalies who played this season, so only the difference from that mix is applied, times the team's shots against per game. By default the mix is re-weighted by this season's starts over goalies not listed out (Out or injured reserve). Teams confirm starters close to game time, so you can pick the starter; day-to-day goalies are shown but not removed.
5. Home ice, rest and travel, measured
Home ice is measured from both seasons for even-strength goals and for power-play chances; neutral-site games get none. A back-to-back is measured on goals scored and allowed. Travel is the distance between arenas since the team's previous game, counted when that game was 1–2 days earlier. A previous game at a neutral site isn't measured. When measuring these effects, the few neutral-site games of past seasons are placed at the listed home team's arena.
6. From expected goals to 10,000 simulations
Expected regulation goals = even-strength goals + power-play goals, then home ice, rest and travel, plus the opposing goalie's difference. The simulator draws 10,000 regulation scores from Poisson distributions. Ties go to overtime/shootout, won by the home side at the share measured on both seasons. It reports win chances, regulation-tie chance, over 5.5 and the most likely scores.
7. What the model doesn't include
Confirmed starting goalies (unless you pick one), skater injuries and line combinations, shot quality or expected goals (xG), time zones, motivation, trades made after the ratings and referees aren't in the numbers. You can edit the expected goals under Custom inputs. Results are estimates, never guarantees. 18+. Bet responsibly.
How our soccer model works
Every soccer simulation, for clubs and national teams, follows the same steps. All numbers come from real completed matches published by ESPN. When there isn't enough data, the simulator says so instead of filling the gap.
1. Real results only
For clubs, the model reads every completed match of the selected competition's current season (Premier League, LaLiga, Champions League and so on). For national teams, it reads each team's last 10 internationals from the past two years in every competition, plus the recent results of every opponent they faced. Extra-time goals are scaled to 90 minutes, and penalty shootouts never count as goals.
2. Attack and defence adjusted for opponents
All results are fitted together to give each team an attack rating (goals scored compared with an average team) and a defence rating (goals allowed). Because the fit knows who each team played, three goals against a strong defence count for more than three against a weak one. A 5-0 against a minnow isn't treated like a 1-0 against a top side. The panel also shows how strong each team's opponents were.
3. Home advantage measured, not assumed
The home edge comes from the same results, comparing home goals with away goals. With few home matches, it is pulled toward no edge. Matches at a neutral venue, like most tournament games, get no home advantage.
4. Careful with small samples
Early in a season, or for teams with few matches, ratings are pulled toward the average as if each team had also played three average matches. International friendlies count half as much as competitive matches, because teams often rotate players. A team that needs more data is flagged, not guessed.
5. From expected goals to 10,000 simulations
Each side's expected goals = competition average × its attack × the opponent's defence × the home edge. The simulator then draws 10,000 scorelines from Poisson distributions to estimate home win, draw, away win, over 2.5 goals, both teams to score and the most common scores. Only regulation time is modeled: there is no extra time, no penalties and no two-leg aggregate.
6. What the model doesn't include
Lineups, injuries, suspensions, expected-goals (xG) data, weather, motivation, travel and fixture congestion aren't in the numbers. Player and lineup information is shown as context only. Each team's goals are drawn independently, so real matches can produce slightly more draws than the model. You can edit any number under Custom inputs. Results are estimates, never guarantees. 18+. Bet responsibly.
Match simulator questions
- Is the Saint Picks Match Simulator available to everyone?
- Anyone can view the page, schedules and matchup information. Running simulations is included with the Saint Picks PRO plan. Premium and the Premium free trial do not include it, PRO has no free trial, and there is no free simulation tier.
- Does the simulator predict the exact final score?
- No. An expected score is a model average, while the animation is one sampled result. Neither is a guaranteed final score or a validated match forecast.
- Do projected lineups and player stats affect the result?
- Team scores still use the displayed team inputs. The player projections tab uses available current-team season statistics and adjusts rates by expected team scoring relative to its baseline. Out and uncertain players are excluded; other projections are conditional on participation. Independent player estimates do not add up to the radar score. Missing data remains unavailable.
- Why does repeating a simulation return the same result?
- A deterministic seed uses the sport, matchup identity and inputs. Under the current model, identical values reproduce the same samples rather than creating a different answer on each click. History is stored locally in this browser, not synced across accounts or devices.
- Can a simulation guarantee a winning bet?
- No. Model uncertainty, missing information and real-match variation remain. These estimates are informational only, for adults aged 18 and over. Bet responsibly.
Assumptions & method
- NFL expected points come from opponent-adjusted red-zone trips, touchdown conversion and other scoring, with home field, quarterback, wind and score spreads measured on real results (see How our NFL model works); regulation ties go to a simplified overtime field goal, and the playback splits the score into touchdowns and field goals as an illustration.
- Each simulation samples goals/points from statistical distributions (Poisson for soccer & NHL, normal for NBA, game-by-game for tennis) driven only by the numbers you enter.
- Soccer rates attack and defence from real results adjusted for opponent strength, with a home edge measured from the same results (none at neutral venues); see How our soccer model works.
- NBA games picked from the schedule use our NBA model: home court, back-to-back effect, team splits, listed absences and margin/total spreads are measured from real results (see How our NBA model works).
- NHL games picked from the schedule use our NHL model: even-strength ratings, power play vs penalty kill, goalies, home ice, back-to-backs, travel and the home share of OT/shootout wins are measured from official NHL results (see How our NHL model works). NHL runs with plain season averages use a fixed 12% home goals edge and an assumed 53% home OT/shootout chance.
- NBA runs with plain season averages (custom inputs or when the model lacks data) use a fixed 2.5-point home edge and 12/16-point margin/total standard deviations; extreme inputs may be unreliable.
- No injuries, lineups, weather or news are included unless you account for them in the numbers yourself.
More free betting tools
Simulations are statistical estimates, not predictions or guarantees. Variance is real: a 70% favorite still loses roughly 3 times in 10. For information only. 18+. Bet responsibly.