Models
Nine models, all public. Want one built for your own question? Work with us. Every Ghost Screen model is built from public data, tested against bars written down before the test was run, and published with the tests it failed. These are Ghost Screen Analytics creations, not NBA stats.
2026-27 Season Forecast
Player ratings from every lineup since 2010-11 become team strengths, and the season is simulated 1,000,000 times a day: records, play-in, playoffs, the lottery and the awards, with the 65-game rule built in.
- Loading test record…
Three Gates: rim protection
A shot at the rim has to get through three gates: Redirect (the shots teams settle for away from the rim), Deter (how rarely they even try at the rim) and Contest (how often rim shots miss). Every defender graded on each, with ranges, not ranks.
- Rim score: teammate independencePass
- Rim score: next-season team forecastPass
- Rim score: season-to-season repeatabilityFail
- Total defense (exploratory): teammate independenceFail
Load & Durability
Who is likely to miss games to injury next season, a body map for every player from the NBA's official injury reports, and which kinds of load actually raise the risk.
- Loading test record…
College to NBA
Every 2026 draft pick and the 2027 draft board, scored from college numbers and draft slot, each with an 80% range.
Open the projectionsBreakout Model
Which young players are likely to be better next season, with an 80% range for each. Five of seven pre-set bars passed; the two misses are shown.
Open the projectionsAging Curve
How fast players decline after their mid-20s, and a 2026-27 projection with an 80% range for every player 26 or older. Six of seven pre-set bars passed; the miss is shown.
Open the projectionsLineup Chemistry
Do five players together beat the sum of their parts? Every lineup's expected and actual net rating, with an 80% range. Four of seven pre-set bars passed: the ranges are honest and familiarity helps, but the model does not beat simply adding up the parts.
- Beats the sum-of-parts line on accuracyFail
- 80% ranges cover 75-85% of lineupsPass
- Real gain beats 95% of shufflesPass
- Returning lineups beat their parts by 0.5+Pass
Franchise Index
Every NBA franchise since 1991, ranked on who built it and who won it. Team reports, season-by-season rosters and moves, and sliders to rebuild the ranking your own way, plus a backtest that predicts each team's point differential from 2004 to 2026.
Open the indexJoy Rankings
The most fun NBA games since 1996-97: Game Score plus a point for every three, steal, block and dunk. Don't like the formula? Move the sliders and rebuild the top 100 yourself.
Open the rankings- A pass/fail bar is written down before each test is run, so the result cannot be bent afterward.
- Only what clears its bar goes into a ranking or a projection. The rest is shown as a description and labeled.
- Every number comes with a range, and every result says what it passed or failed.