Inspired by the original R + lpSolve course project

Build the lineup. See every tradeoff.

Choose a game plan, set real constraints, lock your must-play players, and let an exact optimizer compare every eligible combination.

Open the optimizer

Choose a team-season
Set your game plan
Build and explain the best group

Basketball Reference historical database

Choose a historical team and season

Select any imported NBA team-season from 1980 forward. If a player was traded, only that player's games with the selected team count. The roster is cached on this device.

NBA team Season Season phase

Preparing the imported NBA team-season data...

Current roster data Loading player data... Preparing verified team and player media...

Season: -
Team: -
Players: -

Set the game plan

What should this group do best?

Plan format Choose one five-player group or a larger roster with an exact 240-minute plan. Players in plan Five in lineup mode; eight to twelve in rotation mode. Results to compare Show the best result plus the next-best feasible groups.

Adjust strategy weights

Weights do not need to add to 100. Only their proportions matter. Each stat is converted to a pool-relative percentile before the weights are combined.

Scoring 0 Effective FG% 0 Three-point % 0 Rebounding 0 Playmaking 0 Steals 0 Blocks 0 Ball security 0

Analytics and reporting assumptions

These options explain the result; they do not weaken the exact constraints. Per 36 puts counting stats on an equal playing-time scale. The per-100 view is labelled an estimate because historical team possessions are reconstructed from available team totals.

Stat display Rotation scoring basis

Per game, per 36, and per-100 estimates all describe the selected team stint. Era-relative index compares a player's per-36 rate with the same NBA season and phase; 100 is the minute-weighted league baseline.

In rotation mode, the recommended per-36 basis decides which rate profiles earn minutes. Exact projected totals always use each player's historical per-minute production times the proposed minutes.

Experimental matchup emphasis

Historical Style Counter

Compare two teams from the same season and phase. This transparent heuristic highlights historical statistical gaps and previews an optional weight adjustment; it does not simulate a game or predict a winner.

Opponent

Load a database team-season above to compare an opponent.

Define the rules

Choose who qualifies and what the plan must include

Position coverage

Guards Forwards Centers

Players can cover every source-listed role. For example, a PF-C may fill a forward or center requirement, but each selected player fills only one required roster slot.

Player eligibility

Minimum games Minimum MPG

Only players meeting both requirements can be selected. Recommended defaults are 20 regular-season games and 3 playoff games.

Combined player profiles (optional requirements)

Points ≥ Rebounds ≥ Assists ≥ Steals ≥ Blocks ≥ Turnovers ≤

In lineup mode, these rules add each selected player's historical per-game line. They describe the five-player profile; they do not forecast one team box score.

Choose your players

Lock, exclude, compare, or watch players

Search changes only the visible rows. Eligibility filters apply when optimizing.
Player	Pos	MPG	PTS	REB	AST	STL	BLK	TOV	Lock	Exclude	Compare	Watch
Loading player data...

Loading eligibility...

Best 5-player group

Your result will appear here

Pick a strategy, set any must-have rules, and run the optimizer. You will get a best group, alternatives, and a constraint check.

Transparent by design

Data, scoring, and limitations

The optimizer keeps its math visible and its historical data boundary explicit.

Traceable Basketball Reference data

Team-season totals come from Basketball Reference through the DJHC historical database and are converted to per-game values before optimization. The original 15-player course snapshot and compatible CSV imports remain available as fallbacks.

Pool-relative fit scoring

Each metric is converted to a 0–100 percentile within the eligible, non-excluded pool, then combined using your weights. The result is a pool-relative fit score—not a win probability or universal player rating. Lower turnovers receive a better ball-security score.

Exact constraints

The solver checks every eligible combination within a safe browser limit. Locked and excluded players, position assignments, production floors, and turnover ceilings are hard rules.

Honest limitations

Team-season rows show players who appeared for that team; they are not a current depth chart or injury report. Box scores do not directly measure chemistry, matchups, role changes, or on-court possessions. Any per-100 view is visibly labelled as a team-total-based estimate.

Rotation mode evaluates each candidate with an exact 240-minute plan under your minimum, maximum, and on-court position limits. By default, it ranks counting-stat quality per 36 minutes before allocating time, so historical playing time does not automatically earn more proposed minutes. Projected production still uses each player's historical per-minute production times the exact minutes assigned. It is a historical-stat scenario—not a game prediction.

Historical data boundary

The browser reads a scoped Supabase player pool; it does not scrape Basketball Reference. Each player is limited to the selected team stint, so a midseason trade never adds that player’s all-team aggregate to a lineup. The tool is for historical exploration, not a current roster, injury, prediction, or betting service.