A complete breakdown of the math behind our tournament projections, seedings, and bracketology.
Every game a team plays gets classified into one of four quality tiers based on where the game was played and the opponent's NET ranking. This is the exact system the NCAA Selection Committee uses.
| Home | Neutral | Away | |
|---|---|---|---|
| Quad 1 | vs NET 1-30 | vs NET 1-50 | vs NET 1-75 |
| Quad 2 | vs NET 31-75 | vs NET 51-100 | vs NET 76-135 |
| Quad 3 | vs NET 76-160 | vs NET 101-200 | vs NET 136-240 |
| Quad 4 | vs NET 161+ | vs NET 201+ | vs NET 241+ |
Location matters. A road win at NET #50 is a Quad 1 victory. That same opponent at home is only Quad 2.
Our SOS is computed directly from each team's full-season schedule, every game, every opponent. It blends two independent signals into a single composite score:
| Signal | Weight | What It Measures |
|---|---|---|
| Avg Opponent NET | 65% | Mean final-season NET rank of every opponent played. Uses each opponent’s end-of-year NET (not their ranking at game time) so the schedule is evaluated by how strong opponents turned out to be. |
| Avg Opponent Win% | 35% | Mean winning percentage of all opponents. A team that faced .700 opponents had it harder than one facing .500 opponents, even at similar NET averages. |
Lower composite = harder schedule. Teams are ranked 1-365 by this score and assigned tier labels:
| Tier | Rank Range | Meaning |
|---|---|---|
| Brutal | 1-25 | The toughest schedules in D1. Dominated by Power 5 conference teams in leagues that cannibalize each other. |
| Tough | 26-75 | Above-average difficulty. Mix of mid-tier Power 5 and top mid-major teams. |
| Standard | 76-200 | Average. The broad middle of D1 scheduling. |
| Soft | 201-365 | Below-average. Mostly low-major conferences with few ranked opponents. |
Non-D1 opponents (D2/D3/NAIA) are assigned a proxy NET of 350. Teams without full schedule data fall back to a quad-tier estimation using the midpoint NET of each quad (Q1 ≈ 30, Q2 ≈ 90, Q3 ≈ 170, Q4 ≈ 280).
This blends four major ranking systems into a single number, then adjusts for resume quality. It powers tournament chance (not seeding, that's separate).
The weighting adapts when data sources are missing:
| Scenario | NET | KenPom | Barttorvik | Sagarin |
|---|---|---|---|---|
| All available | 30% | 30% | 25% | 15% |
| No Barttorvik | 35% | 35% | n/a | 30% |
| No KenPom | 35% | n/a | 30% | 35% |
| NET + Sagarin only | 55% | n/a | n/a | 45% |
Five factors shift the composite up or down:
This is a completely separate calculation from the composite power rating. It models how the NCAA Selection Committee actually evaluates teams for seeding, based on committee member public statements and historical bracket analysis (2018-2025).
Six weighted factors, each producing a score where lower = better seed:
| Factor | Weight | What It Measures |
|---|---|---|
| SOS-Adjusted NET | 35% | NET ranking amplified or dampened by schedule strength. Hard schedule makes NET more trustworthy; soft schedule suggests inflation. Adjustment capped at ±15%. |
| Quad Quality Index | 25% | Game-by-game resume score. Q1 wins have diminishing returns (the 1st Q1 win is worth more than the 15th). Q3 losses cost −5, Q4 losses cost −10. |
| Q1 Win Power | 15% | Uses Bayesian regression to prevent small-sample flukes. A team that's 2-0 in Q1 isn't truly 100% Q1 quality, we add 6 pseudo-games at 50% to keep it honest. Also rewards volume. |
| Strength of Schedule | 10% | Raw SOS rank. Harder schedule = better seed. |
| Conference Tier | 10% | Historical bias: Power 5 teams seed 3-5 spots higher than mid-majors with identical profiles. |
| Eye Test | 5% | AP Poll ranking + win percentage as a proxy for national perception. |
Tournament chance measures one thing: how likely a team was to earn an at-large bid on résumé alone.
All 365 Division I teams were ranked by composite power rating. A sigmoid function converts rank into probability:
The midpoint was set dynamically to the overall rank of the last at-large team to make the field (68 teams through 2025-26, 76 from 2026-27). Teams well above the cutoff approach 99%. Teams right at the cutoff land near 50%. Below the cutoff, probability drops sharply.
Conference-tournament champions entered the field via their automatic bid regardless of this number, they're shown with a separate AUTO-BID badge, and their percentage remains the at-large-only chance.
The field is built in five steps. Its size depends on the season being shown: the NCAA tournament expands from 68 teams to 76 beginning with the 2026-27 season, and the page follows whichever season the data describes rather than a fixed number.
| Through 2025-26 | 2026-27 onward | |
|---|---|---|
| Field | 68 | 76 |
| Automatic bids | 31 (one per conference) | 32 |
| At-large bids | 37 | 44 |
| Play-in round | First Four: 4 games, 8 teams | Opening Round: 12 games, 24 teams |
| Teams remaining after it | 64 | 64 |
The NCAA Selection Committee uses a "snake" pattern to distribute teams across four regions so that no single region is dramatically stronger than another.
This ensures the overall #1 team faces the weakest #2 seed in their region, while the #4 overall faces the strongest #2 seed. Regions stay balanced.
Each of the four play-in games is assigned to a region using the tail positions of the S-curve:
Every region hosts exactly one play-in game. Both teams in a play-in share the same seed and region, the winner advances into that bracket slot.
We surface four groups of four teams each to give a clear picture of the bubble:
| Group | Who They Are |
|---|---|
| Last Four Byes | The 4 at-large teams just above the First Four cut. They made the field without having to play in. |
| Last Four In | The 4 at-large teams in the First Four play-in games at the #11 seed. They're in, but barely. |
| First Four Out | The best 4 teams that just missed the field entirely. |
| Next Four Out | The next 4 closest misses after that. |
The Bracketverse Rating is our single composite number for every Division I team. It blends three independent systems into one percentile-normalized score from 1 to 99, where higher is better.
| Component | Weight | Source |
|---|---|---|
| Power Rank | 50%* | Composite blend of NET, KenPom, Barttorvik, and Sagarin (Section 3), adjusted for quad record, SOS, conference, streak, and AP ranking. Captures how strong a team is. |
| Seed Score | 30%* | Committee-style 6-factor resume model (Section 4). Captures how the selection committee would evaluate a team: SOS-adjusted NET, quad quality, Q1 win power, schedule strength, conference tier, eye test. |
| Trajectory | 20%* | Win/loss streaks, AP Poll movement, and ranked status. Captures which direction a team is heading. A team on a 10-game win streak is trending differently than one that just lost 3 in a row, even if their season numbers look similar. |
*Early in the season (before streak data is available), the weights automatically shift to 60% Power Rank / 40% Seed Score. Once trajectory data lands, the full 50/30/20 blend activates. No manual adjustment needed, the system adapts. The full 50/30/20 blend is currently active.
Michigan is the cleanest possible case: best in all three components, so the blend is a flat 1.0 and no weighting argument can change the answer. The contrast is more instructive. Duke finished 35-3 and ranked #3 in both power and resume, but lost its last game. An L1 streak is worth exactly zero (not a penalty, zero) because a single result is noise in either direction. Duke's whole trajectory score is its AP bonus, (26−4)×0.3 = +6.6, which places it #6 on the trajectory board rather than near the top. That is the difference between finishing #1 and finishing #3 in a season where the two teams were a rounding error apart everywhere else.
Our own weekly-style ranking of Division I teams, powered entirely by the Bracketverse Rating. No voter bias, no ballot stuffing, no regional favoritism. Pure data.
Each team gets a tier label based on their Bracketverse Rating value:
| Tier | Rating Range | What It Means |
|---|---|---|
| Elite | 97-99 | Championship-caliber teams. Dominant across all metrics, power, resume, and trajectory. |
| Strong | 93-96 | Clear tournament teams with strong resumes. Likely seeded 3-6 in bracketology. |
| Solid | 88-92 | Tournament-quality teams with some resume gaps. Typically seeded 7-11. |
| Bubble | Below 88 | On the bubble or just outside. Teams that needed late-season wins to secure a bid. |
There are no rules like “must include at least one team from every power conference.” If 8 of the top 10 are from the SEC, that’s what the data says. We let the numbers speak.
The Most Cursed board ranks the fanbases who have suffered most in the last ten seasons. It is deliberately not a measure of how good a program is, the teams at the top are mostly very good, because it takes a contender to lose in the ways that actually hurt.
Every season inside the ten-year window can contribute at most one moment, so a single catastrophic year cannot be counted twice. Moments come from two places:
| Source | What it gives |
|---|---|
| Hand-written history (54 programs) | A specific, fact-checked line for the seasons that defined a fanbase's decade: “31-0 and unbeaten, then blown out by Baylor in the title game.” Where a curated moment exists for a season, it wins. |
| The tournament feed (all 365 teams) | Every tournament appearance from 2016 on, scored automatically from its seed and how far it got. This fills in every season and every program the hand-written list does not cover. |
An automatically-scored moment is graded on how far short of its seed the run ended, then aged. A 1-or-2 seed is expected to reach the Elite Eight, a 3-or-4 seed the Sweet 16, a 5-to-8 seed the round of 32. A 9-seed or lower is expected only to make the main draw, so losing its opening game costs nothing, though being knocked out in the First Four, before the bracket proper, still registers.
Then recency: a moment keeps between half and all of its weight depending on how long ago it was, so last spring stings more than 2017 without erasing it.
A fanbase's moments are sorted worst-first, and the worst one dominates. Each additional moment still counts, but a little less than the one above it, enough that a decade of repeated near-misses outranks a single bad night, without letting a long list of mild disappointments outweigh a genuine catastrophe.
| Program | Last title | Where it lands |
|---|---|---|
| Florida | 2025 | Off the board entirely |
| UConn | 2024 (and 2023) | Off the board entirely |
| Kansas | 2022 | Still on it, around 34th |
| North Carolina | 2017 | Still on it, around 26th |
The board is scoped to 88 programs: the ACC, Big Ten, Big 12, SEC and Big East, plus nine mid-major bluebloods with a genuine national record: Gonzaga, Memphis, Wichita State, San Diego State, Saint Mary's, Dayton, VCU, Loyola Chicago and Florida Atlantic. A team also needs at least two real moments in the window; one bad night is not a curse. Sixty-one programs currently clear that bar, and the board shows the 50 most tortured.
Every board team carries a label: Cursed, Snakebit, Star-crossed or Hard-luck, assigned by where a team sits on the board, not by an absolute score: roughly the top 15% are Cursed, the next 30% Snakebit, the next 30% Star-crossed, the rest Hard-luck.
The underlying number is deliberately never shown. It is an ordering device, not a measurement, there is no honest way to say one fanbase's decade was “73 units” of pain. The rank and the tier are what the number is good for, so they are all the board publishes. Each team's page shows the same tier alongside the actual moments behind it, which is the part that means anything.
Data sources: the hand-written moments for 54 programs, plus the tournament-history feed covering 211 programs and 627 tournament appearances (2016-2026). The board is recomputed whenever the underlying data refreshes, and the ten-year window moves with the calendar, a 2016 heartbreak drops out of the reckoning after this season.
Every other ranking on this site is computed from results. The People's 25 is the one ranking you write. Each week the board opens for voting: for teams 2 through 25 you say Higher, Same or Lower, and you may nominate up to three teams you think belong. The team at #1 is asked a different question (there is nowhere above it), so the choice there is Keep at #1 or Lower. When the window closes, every ballot is composited into next week's board.
A team's movement comes from the gap between its Higher and Lower votes, measured against everyone who voted this week, not just the people who happened to rate that team. Ignoring a team is therefore not the same as backing it.
The + 4 is a cold-start brake. Without it, a single voter in a quiet opening week would move a team as far as five hundred voters could. With it, that first lone ballot moves nothing, and the brake has faded to nothing by the time twenty people have voted.
| Vote split on a team ranked 13th | Where it lands |
|---|---|
| Every voter says Higher | 9th (+4) |
| 70% Higher / 20% Same / 10% Lower | 11th (+2) |
| 60% Higher / 40% Lower | 13th (no move) |
| Every voter says Lower | 17th (−4) |
Measured on a full 25-team board with 500 voters, and symmetric in both directions. The exact landing spot also depends on how a team's neighbours moved, everyone shifts at once. A realistic week with 200 voters and genuinely mixed opinion moves the biggest riser two spots.
A team outside the board enters when at least 15% of ballots nominate it, and never on fewer than 10 ballots. Everyone gets three nomination slots, so scattered picks spread thin; fifteen percent landing on one name is a real signal rather than noise. The ten-ballot floor stops a quiet opening week from seating a team on three votes.
Newcomers debut near the bottom, and how high depends on how many people asked for them: 15% of ballots seats a team 25th, 30% seats it 23rd, and anything from 60% up seats it 21st, which is the ceiling. A poll seats new arrivals at the back and lets them climb.
The composite is covered by a contract test that checks the properties any honest poll must have: a team everyone pushes up never falls, one voter in a hundred cannot move a team more than a spot, ballot order is irrelevant, malformed ballots are ignored rather than fatal, and the same ballots always return the same board.
The same 1 to 99 scale we give teams, applied to individual players. A 99 is the best player in the country by this measure, a 50 is exactly average among players who qualify, and a 1 is the bottom of the pool. It is built from the box score and nothing else, which is both its strength and, as set out at the bottom of this section, its limit.
A player needs at least 10 games and at least 10 minutes per game. That cut exists because rate statistics are meaningless on tiny samples: a walk-on who plays four minutes and hits his only shot would otherwise post a perfect field-goal percentage and outrank an All-American. Players appearing on two rosters, which is what a transfer looks like in the data, are counted once.
2,052 players qualify today. Everyone else on a roster is still on the site, they are simply not ranked.
Each component is scored by rank across the whole pool, converted to a percentile, then weighted. Ranking first, rather than blending the raw numbers, keeps a statistic with a big natural spread from drowning out one with a small spread.
| Component | Weight | What it is, and why |
|---|---|---|
| Scoring impact | 30% | Points per game multiplied by field-goal percentage over 45. It rewards scoring and efficiency together, so volume taken at a bad rate does not pay. Dividing by 45, a little below the Division I average, means an average shooter is left roughly where he started and only genuinely efficient scorers gain. |
| All-around production | 25% | Rebounds, assists, steals and blocks added together. This is the part of the game that never shows up in a scoring average, and it is weighted second-heaviest on purpose so that a distributor or a rim protector can rank highly without scoring much. |
| Scoring volume | 20% | Raw points per game, with no efficiency adjustment. Carrying a large scoring load is a real burden and a real skill, and this is the component that credits it plainly. |
| Shooting efficiency | 15% | Half field-goal percentage, a quarter free-throw percentage, a quarter three-point percentage. Field-goal percentage carries the most weight because it applies to every shot taken; the other two are narrower skills. |
| Team strength | 10% | The NET ranking of the team the player is on. A small thumb on the scale for the fact that 18 points against elite defences most nights is harder than 18 against weak ones. Deliberately the smallest weight, so a great player on a bad team is not buried. |
36 games, 33.5 minutes, 19.5 points, 3.6 rebounds, 5.1 assists, 2.4 steals, 0.3 blocks, shooting 48.5% from the field, 36.8% from three and 85.3% from the line.
| Component | His rank | Percentile | Weighted |
|---|---|---|---|
| Scoring impact 19.5 × (48.5 ÷ 45) = 21.0 | #27 of 2,052 | 0.987 | 0.296 |
| All-around 3.6 + 5.1 + 2.4 + 0.3 = 11.4 | #73 | 0.965 | 0.241 |
| Scoring volume 19.5 PPG | #24 | 0.989 | 0.198 |
| Shooting efficiency 54.8 | #150 | 0.927 | 0.139 |
| Team strength Vanderbilt, NET 15 | #105 | 0.949 | 0.095 |
| Blended score | 0.969 | ||
| Rank, and rating | #1 of 2,052 → 99 | ||
Players carry the same three labels as teams, at the same cut-offs, because both numbers are percentiles on the same 1 to 99 scale. Using one set of thresholds for both boards is what makes the words mean the same thing in each place.
| Tier | Rating | Players | Teams |
|---|---|---|---|
| Elite | 97-99 | 53 of 2,052 (2.6%) | 13 of 365 (3.6%) |
| Strong | 93-96 | 84 (4.1%) | 15 (4.1%) |
| Solid | 88-92 | 104 (5.1%) | 18 (4.9%) |
The proportions land in much the same place on both boards, which is the test of whether shared thresholds were the right choice. Elite is the top 2.6% of ranked players and the top 3.6% of teams; Strong and Solid match to within a tenth of a point. A label means roughly the same rarity whichever board you are reading.
Note that the tiers describe where a player sits among the 2,052 who qualify, not among everyone in Division I. More than 5,500 players appear on a roster, so an Elite player is inside the top 53 of those who played enough to be measured at all.
Three things are worth knowing before reading much into a player's rating.
The deeper limit is that this is a box score, so it sees what the box score sees. Defence is only steals and blocks, which misses nearly all of it. There is no adjustment for the quality of the shots a player takes, for how much of his offence he creates himself, or for the strength of the specific opponents he faced. Efficiency metrics that attempt those things need play-by-play data we do not currently carry. When we do, this section will change and will say so here.
The regular Bracketverse Rating is built from results: record, quad wins, strength of schedule, momentum. Before the first tip-off none of that exists. Run on an empty season it would simply hand back last year's order, and a team that lost its whole starting five would sit exactly where it finished. So before the season we rate teams a different way, from who is actually on the roster.
It uses the same 1 to 99 scale and the same Elite, Strong and Solid cut-offs, so a preseason 97 is as rare as a March 97. It is a different measurement, though, and is always labelled Preseason.
Each player on a 2026-27 roster is valued from the best evidence available for him:
| Player | Evidence | How he is valued |
|---|---|---|
| Played in Division I last season returners and transfers | Last season's box score | His production, discounted by the level of competition where he earned it |
| Ranked recruit 247Sports Composite top 150 | His national ranking | What last season's freshman of the same rank actually produced |
| Everyone else unranked freshmen, junior college and overseas arrivals, walk-ons | None | A flat bench value |
Production is a simplified version of the Game Score statistic: points, rebounds, assists, steals and blocks, minus turnovers, with scoring nudged up or down by shooting percentage.
Level of competition matters. Twenty points a game in a weak conference is not twenty in the SEC. Production earned at the country's best team counts in full; at the weakest, it counts 55%, with every team in between on a straight line by its NET ranking. Players who appeared in only a handful of games are pulled toward the bench value, because a few good nights are not proof.
Recruits are priced by history, not by stars. We ranked last season's freshmen by what they produced and found the top of that list was exactly the top of last year's recruiting class. So this year's #1 recruit is valued like last year's most productive freshman, the #50 like the fiftieth, pulled a quarter of the way toward the middle because recruiting rankings are good but not perfect.
A team is its best eight players, weighted by how much each role typically plays: the five starters count most, the bench less. A roster with fewer than eight valued players is filled out with bench value, so a thin roster is marked down rather than ignored.
Programs also carry something beyond any one roster: coaching, system, player development. So the final rating is 80% roster and 20% last season's final Bracketverse Rating. When a program has changed head coach, that drops to 10%, because much of what "program" means leaves with the coach.
Kansas brings back only a sliver of last season's minutes. Rank it by last year's results and it looks like a team that finished 18th. Rank it by who is on the roster and it looks very different.
| Player | Where the evidence comes from | Value |
|---|---|---|
| Tyran Stokes | #1 recruit in the country | 26.6 |
| Taylen Kinney | #22 recruit | 16.4 |
| Leroy Blyden Jr. | 16.4 points a game at Toledo (NET 134) | 16.3 |
| Keanu Dawes | 12.5 points a game at Utah (NET 133) | 13.7 |
| Dennis Parker Jr. | 18.3 points a game at Radford (NET 252) | 13.7 |
| Christian Reeves | 11.1 points a game at Charleston (NET 165) | 12.7 |
| Davion Adkins | #72 recruit | 12.0 |
| Trent Perry | #105 recruit | 10.9 |
Notice Dennis Parker Jr. He scored more than anyone else in that list, but he did it at Radford, so he is valued below Leroy Blyden Jr., who scored less against tougher opposition at Toledo. That is the level adjustment doing its job.
Those eight make Kansas's roster the fifth-best in the country. Blended with a strong program score from last season, Kansas is rated 4th, a 99.
The Preseason Bracketverse Top 25 is simply the top of this board: the 25 highest preseason ratings, with the next ten listed as others receiving votes, exactly as the regular Top 25 does during the season. It is frozen along with the ratings, so it stays as it was published and can be checked against how the season actually goes.
Each team is shown with where it finished last season, because the movement is the interesting part. Some teams climb a long way on the strength of who arrived, others fall because of who left, and the defending national champion can land well below first if its coach and best players have moved on.