Rankings / Methodology

Rating Definitions

2025–26 men’s Division I. 365 teams; 5,737 accepted games through 2026-04-06.

ORtg

100 × total points scored ÷ total estimated possessions. Higher is better.

DRtg

100 × total points allowed ÷ total estimated possessions. Lower is better.

Net

ORtg − DRtg. Default ranking is descending net efficiency.

Tempo

40 × total estimated possessions ÷ game minutes, including overtime.

Possessions

FGA − offensive rebounds + turnovers + 0.475 × FTA, averaged between both teams for each game.

D-I W–L

Record in accepted games against teams in the official NCAA Division I season directory.

Season ratings use sums of points and possessions, rather than an unweighted average of game ratings. Display rounds to one decimal place; ranking uses full precision.

These are raw efficiencies. No adjustment for opponent strength, venue, recency, or preseason expectations has been applied. They are not KenPom ratings.

Experimental model estimates

The home-page rankings table includes independent cutoff estimates for all 365 teams. The default row order remains raw net efficiency. These estimates have not passed every predictive calibration, subgroup and uncertainty gate; they are not KenPom or official NCAA metrics.

AdjO / AdjD

Possession-weighted additive offense and defensive allowance, adjusted for opponents and explicit home/away/neutral designation. Ridge regularization shrinks sparse team effects. Both are centered on the fitted Division I baseline; lower AdjD is better.

AdjNet

AdjO − AdjD. Model-view ranks use this full-precision value.

AdjT

Opponent-adjusted possessions per 40 minutes. An additive pace model weights games by minutes, includes overtime and centers team effects. Faster is a style, not inherently better.

SOS

Mean opponent cutoff AdjNet minus twice the fitted efficiency home effect times own venue sign (+1 home, −1 away, 0 neutral). All accepted Division I games are included; higher means harder. This is our transparent schedule-strength definition.

NCSOS

The same formula applied only to games explicitly designated regular-season nonconference in the corroborated frozen ESPN schedule. Conference tournaments and national postseason are excluded. It is not inferred from team names or the current conference.

Expected wins

Sum of cutoff matchup probabilities against the completed schedule. This is retrospective expected record, not a record of pregame predictions. Shown on team pages.

RR

Actual accepted wins minus cutoff expected wins. Positive residual is not proof of luck; it is a descriptive difference under this model.

Model WAB

Actual accepted wins minus expected wins for a fixed cutoff benchmark against the same schedule at the same venues. The benchmark is the 45th team by this model’s adjusted net: Villanova. This independent model WAB does not reproduce official NCAA NET-45 WAB.

Cutoff probability calibration uses the four prior seasons, with a development-selected blend of raw and adjusted matchup strength that gives raw information greater weight in smaller samples. All pregame historical comparisons exclude target-date results. The current model uses completed-season ratings for schedule summaries; it must not be mistaken for a contemporaneous forecast. Undefined venue-sensitive totals stay unavailable rather than treating missing venues as neutral. 5,736 of 5,737 venues are reconciled. Penn–Villanova has conflicting institutional designations, so both teams’ complete schedule-derived totals are withheld.

Team page definitions

Min%

A player’s logged minutes divided by the team’s game minutes (40 per game plus five per overtime). A player on court for every minute reads 100%; the five slots sum to 500%.

Effective rotation size

Derived, not published. Each player’s share of all logged minutes is squared and summed; the effective size is one divided by that sum. Ten players with equal minutes read 10; one player with all minutes reads 1.

Top three and bench share

Derived from the same shares. Top three is the share of logged minutes played by the three players with the most minutes. Bench is the share played by everyone outside the five players with the most minutes. No positions are assigned.

Sample flags

Descriptive thresholds, not model output. Few games: under 10 logged games. Under 100 minutes: under 100 logged minutes. A dagger marks a rate on under 100 field-goal attempts (2P%, eFG%, TS%, FTA/FGA), under 50 three-point attempts (3P%) or under 25 free-throw attempts (FT%).

Last five

The team’s five most recent games. The small figure beside a last-five value is its difference from the season value, shown only when at least one of those games has a player log.

Rank mark

A small mark beside each national scouting rank. Its horizontal position repeats the printed rank on a scale of 365 teams, with the best rank at the right. It supplements the printed rank and the cell shading.

Player value by season

The player page lists every season in which the model produced a value with a range, newest first. Each row prints net, offense, defense, a range (10th to 90th percentile of 200 whole-game resampling draws, not a calibrated interval), an evidence label (sufficient: 600 or more on-court possessions; limited: 300 to 599), the split role of the season, and wins above replacement where it was computed (2023 onward). Split role: training and development seasons were used to fit or select the model, evaluation seasons were examined post hoc, and the descriptive season was not used in fit or gates. Each season is a separate retrospective fit, so no change or trend is shown, and the table is ordered by season only, never by value. Seasons before 2018 have a point value in the model files but no range, so no number is shown; a season the player played with no published value is listed as its own row with its split role and the reason. Flags printed beside a season: season-2021 marks the 2020-21 season, shortened by COVID-19 cancellations and kept in the model fit; unbound-teammates-share marks a player for whom more than 5% of on-court possessions included at least one of the other nine players on court (teammate or opponent) not matched to the player registry; prior-from-box-only marks a prior built from the box-score line without a prior-season value.

Team totals

Season sums of the team box score and of the opposing teams’ boxes in the same games. Observed arithmetic, not a model value, unadjusted for opponent or pace. Steals and personal fouls have no team figure in the source, so only the sum of player lines exists. Offensive and defensive rebounds exist for the team only, because player lines carry combined rebounds. Team minutes are derived (five players times game minutes), not observed.

Per game, per 40 and totals

Per game is the total divided by games played. Per 40 is the total times 40 divided by game minutes (40 per game plus 5 per overtime period). Totals are season sums. The basis chosen above the roster table also sets the basis of the team totals table.

Residual, tolerance and status

Residual is the team total in games with player logs minus the sum of player lines in those games. It is always shown; it is never dropped for being large. Tolerances were declared before the data were built and were not changed afterwards: counting statistics reconcile within 2 or 0.5% of the logged total, turnovers within 5 or 3%, rebounds within 10 or 8% (whichever is larger). Status: reconciled is within tolerance, residual shown is outside it, players exceed team means player lines add to more than the team total (a data problem), no team figure and no player split mean the comparison cannot be made. Team rebounds and team turnovers are credited to the team and not to a player, so most teams show a residual for those two statistics. A check declared in docs/team-residual-check-criteria.json found every team’s residual for them at zero or above, with medians of 3.4 rebounds and 0.6 turnovers per logged game, which is consistent with that explanation and does not prove it.

Team aggregate of player values

Experimental and model-implied. It combines the season’s player values with their share of on-court time, counting players under 300 possessions at the replacement level. The headline is the average over the five on-court slots; the five-slot sum is shown too, and a sum such as +70 is an artifact of adding five slots on the player-value scale, not a team margin. Its range comes from joint resampling draws, so it is not the sum of the players’ separate ranges. The wins figure applies the win conversion to each player separately, so it can exceed games played. It is not comparable to AdjNet or the record, and no list ranks or orders teams by it. Every value in it is a 2026 value, so one split role describes it (descriptive: not used in fit or gates).

Player value strip

Shows the published model-implied net value and its range for each player, in possession-share order. The graphic uses a shared axis with a line at zero. Open circle: limited evidence. Filled circle: sufficient evidence. Nothing is drawn when evidence is insufficient. It is not sortable, and no list ranks players by value. The model is experimental and not promoted; see the NIL section below for failed gates and the post-hoc disclosure.

Team identities and color

Conference membership belongs to each team season and follows the Kaggle historical conference records. Missing membership stays unavailable. Team swatches use frozen NCAA.com colors, supplemented by ESPN and documented school brand sources. Historical pages use current palettes, not reconstructed historical colors. Wins are green and losses red. Ranked metrics use a relative scale from green at the best rank through yellow at the median to red at the worst rank. Rankings already account for whether higher or lower values are better. W/L and numeric ranks remain visible.

Team scouting and player reports

Scouting values use summed accepted team boxes. Effective FG% = 100 × (FGM + 0.5 × 3PM) / FGA; turnover rate = 100 × turnovers / estimated possessions; offensive rebound % = 100 × ORB / (ORB + opponent DRB); free-throw attempt rate = 100 × FTA / FGA. Two-point shooting subtracts threes from total field goals. Assist rate uses FGM; block rate uses opponent two-point attempts. Point distribution divides points from each shot type by total points.

Defense columns describe opponents’ corresponding statistics. Division I averages pool both participants in all accepted games, using sums rather than averaging percentages. National ranks use full precision and competition ranking (ties share a rank); lower offense turnover rates and lower defensive shooting/rebounding/efficiency values rank better. Defensive turnover rates rank higher values first. These remain unadjusted statistics.

Team player tables aggregate only appearances for that team. Minute share divides logged player minutes by all team game minutes, including overtime. Per-game values use the player’s logged appearances. The last-five rotation uses the latest five accepted team games, not the player’s five most recent appearances. Missing records are not treated as zero-performance games.

Lineups come from SportsDataverse’s reconstructed on-court player IDs, linked through existing season identities. Accepted intervals require five distinct linked players, clean source status, monotonically increasing elapsed clocks, at most 30 unobserved final seconds, and agreement with player box minutes within 90 seconds per player. Unobserved final seconds are excluded from coverage and lineup minutes. Lineup shares divide observed group minutes by all last-five game minutes. Groups are unordered and positions are not inferred. Native NCAA play-only repairs do not supply reconstructed lineup fields.

Schedule venue labels use uniquely matched Kaggle historical participants with the same linked team identities, date, and both scores; an unmatched venue remains unavailable. Designated home/away score sides alone do not determine venue.

Data Sources

Source Catalog and Captured Artifacts

Team identities come from the direct NCAA 2026 Division I directory. Full-season team statistics are reconstructed in the secondary NCAA-derived SportsDataverse archive. The 160 direct NCAA sample finals all agree with their published season game scores.

Individual player statistics are published across accepted season games when player identities, shooting arithmetic, and summed scoring reconcile to team finals. Play-by-play publication covers 5,716 games and 2,782,378 events whose teams, dates, clocks, and final scores pass validation. Two additional rated games have invalid clocks and are withheld.

Historical team results and raw efficiencies are available in History. Validated NCAA-derived player statistics and play-by-play cover available games from 2010–2025; native ESPN player boxes supplement accepted games from 2004–2009. Linked career pages track corroborated seasons and team associations. Cross-provider player links require repeated matched games, team agreement and exact individual statistics; names alone never establish identity.

Player performance aggregates sum accepted appearances. Per-game rates use logged games; per-40 rates use logged minutes. Shooting percentages use summed makes and attempts. Estimated true shooting uses points ÷ (2 × (FGA + 0.44 × FTA)). Missing games are excluded, and zero-denominator rates are unavailable. Career rates are recomputed from combined counts, rather than averaged across seasons.

Team archive ↗ · Play-by-play archive ↗

Historical Collections · 203,027 games across 42 seasons. Historical efficiency uses available detailed games only.