Comprehensive analysis of NBA Draft performance from 2010-2026. Explore GM rankings, draft classes, and discover which teams consistently find value in the draft.
See which games feature 2026 prospects and when they’re playing. Your daily guide to tomorrow’s NBA stars.
OpenSortable big board with college stats, BPM, and rankings from ESPN, Tankathon, Vecenie, and Mohamed.
OpenComposite rankings and scouting notes from Mohamed (@mcfNBA). Per-source breakdowns and the latest updates.
OpenFull 2026 class board with ratings, tiers, comparisons, and methodology in one place.
OpenData-driven analysis ranking every NBA General Manager by draft success from 2010-2020. Discover which GMs consistently find value, avoid poor picks, and build championship-caliber rosters through the draft.
Dive deep into 17 years of draft analysis with our machine learning-powered rankings. Compare our projections against actual NBA outcomes and see how each draft class stacks up historically.
The top 50 college prospects by projected prime impact rating across every draft class from 2010 to 2026. Who were the best-rated players entering the league each year?
Every NBA Draft class from 2010–2026 ranked by the average rating of their top 35 prospects. Which years produced the deepest talent?
Browse every NBA Draft from 2010-2026. Click on any year to see detailed analysis, player performance, and draft outcomes.
12 draft classes (2010–2021), leave-one-year-out backtest, compared head-to-head against the NBA's actual pick order — same player pool on both sides, no cherry-picking.
targetPM is a player's prime-years two-way NBA impact — a blend of 35% xRAPM (an on/off regularized adjusted plus-minus that captures impact the box score misses) and 65% BPM (box plus-minus), averaged across age 24–28 seasons in which the player logged at least 1,500 minutes. It is an impact metric, not a counting stat: it tests whether we correctly identified the most impactful players, not just the ones who hung around. 391 prospects from 2010–2021 have observed prime-years data, and our model is scored against the NBA's pick order on the exact same pool.
The model projects defensive impact (Spearman 0.59) about 50% more accurately than offensive impact (Spearman 0.40) — and we think that's about as good as anyone can do with pre-draft data. Defense lives in measurables — rebounds, steals, blocks, length — that travel cleanly from college to the NBA. Offense lives in NBA-side context — usage, role, surrounding shooters, scheme — that nobody can fully project before draft night.
The same asymmetry shows up across individual stats: the model is most reliable at projecting the shape of a player's NBA game — how often they shoot threes, rebound, pass, block shots, draw fouls, and make free throws — and less precise at projecting opportunity stats like minutes and usage, which depend heavily on team context, injuries, and development environment.
Leave-one-year-out validation across the 12 draft classes from 2010 to 2021. For each class, the model is trained only on data available before that draft — every prediction for, say, the 2018 class comes from a model that has seen 2010–2017 and 2019–2021 but not 2018 itself. This eliminates the most common form of overfitting (training on the same data you evaluate on).
A player's prime-years impact, defined as a 35% / 65% blend of xRAPM and BPM averaged across NBA seasons in his age 24–28 window, restricted to players who logged at least 1,500 minutes inside that window. xRAPM is a regularized adjusted plus-minus that captures on/off impact the box score misses; BPM is the familiar Basketball-Reference box plus-minus. Blending the two pulls in real on-court signal without letting the noisier RAPM dominate. This is intentionally an impact metric rather than a counting metric — we want to test whether we identified the most impactful players, not just the ones who stuck around. 391 players from the 2010–2021 classes have observed prime-years data.
A familiar Basketball-Reference accumulation stat (Value Over Replacement Player). 580 drafted players have non-zero NBA career data in the sample.
For every comparison against NBA draft order, we use the same pool of players for both sides. Our model and the actual pick are scored on identical samples in identical ways. This is conservative: it deliberately excludes prospects where our model has no prediction, which would otherwise let us cherry-pick a more favorable baseline.
Predictions come from the same Stack PostDraft blend that powers the live draft board: an ElasticNet regression layer (with per-archetype models for guards, wings, forwards, and bigs), a LightGBM boosted-tree layer, and a scout-consensus prior. Anthropometric features (height, wingspan, plus-length, weight) are first-class inputs, with a flag indicating whether a player's measurements were directly recorded or inferred from height-conditioned medians.
Why compare against actual NBA picks rather than pre-draft mock consensus?
The actual NBA draft order is the league's collective pre-draft consensus, weighted by what teams were willing to bet draft capital on. It's the most accountable baseline available. (Our model also folds in a public-consensus prior as one of many features, so we'd be comparing partly against ourselves.)
What about players who never made the NBA?
They're excluded from the primary metric — no observed prime-years impact to score. The secondary VORP metric includes them at zero or near-zero, which mostly tracks correctly: undrafted prospects with no NBA career contribute about equally to both sides' rank error, since neither our model nor the NBA ranked them highly.
Have these numbers been cherry-picked from one model variant?
No — they come from the same blended model that drives the live draft board. We also track ElasticNet-only, LightGBM-only, and stats-only variants internally; the shipping blend is not the most flattering variant on every single metric — it's chosen on overall robustness.
Why 1,500 minutes and ages 24–28 for the prime window?
1,500 minutes is roughly half a season of meaningful playing time — the smallest window where BPM is statistically reliable. 24–28 is the conventional peak age band for NBA players; narrower bands give similar conclusions on a smaller sample, wider bands include early-career years that aren't yet prime impact.
Backtest: 12 draft classes (2010–2021), leave-one-year-out validation, strict matched player pool. Primary metric: targetPM (prime-years ±impact, 391 players). Secondary: career VORP (580 drafted players).
Our ratings project a player's long-term NBA impact — their prime-years two-way contribution, not rookie-year production. Every player receives an Overall Rating, an Offensive Rating, and a Defensive Rating on a 0–100 scale.
Outcomes from the most statistically similar historical prospects, weighted by career totals.
High-end outcomes from within that same comparable player pool — the ceiling scenarios.
Position-aware regression trained on college stats and advanced metrics across 15+ draft classes.
Same regression, but with scout and media consensus rankings folded in as an additional input.
Players without a U.S. college statistical record receive manually set ratings anchored to the consensus draft board at the time and Kevin Pelton's international projections. These entries are excluded from global model calibration so they don't distort the fit for everyone else.
Why projections tend to be conservative: The NBA stat estimates in each player's profile are weighted averages of similar players' full career outcomes — busts included, not just breakouts. They represent a realistic expectation, not a ceiling.