About
The honest all-time NBA team builder
SweepSzn is a free browser game. Spin a franchise reel and an era reel, draft a five-player all-time NBA starting five, and the engine simulates a full 82-game season. The twist: it's honest. Instead of rewarding you for stacking five stat-stuffers, it models the things that actually win an NBA season — and then tells you exactly why your lineup goes 78-4 or 40-42.
The whole hook is the perfect season. Can you build a five that goes 82-0? It's sweep season.
Here's what a real engine result looks like — a balanced two-way GOAT five that still loses wins to usage overload.
Live engine output — the same model that grades your draft.
FAQ
Is SweepSzn free?
Yes. It runs entirely in your browser, with no account required. Optional Google sign-in just lets you claim your leaderboard ranks and keep them across devices.
Can you really go 82-0?
Nobody has. The best draftable five projects 79-3 — better than the 73-9 Warriors, the best real season in NBA history — and it takes a genuinely balanced two-way five, not five ball-dominant scorers. (The engine's theoretical maximum is 80-2, but that lineup needs two centers in one starting five — you can't actually draft it.) The chase is the whole challenge.
How accurate is the engine?
Every coefficient is fit to 1,170 real NBA team-seasons (1985–2025) across 24,687 player-seasons, with out-of-sample accuracy of about 5.6 wins RMSE against luck-adjusted Pythagorean win totals in year-grouped cross-validation (6.1 against raw win totals, which carry roughly 2.4 wins of close-game luck). Wins come from a Pythagorean expectation. It is calibrated to history, not hand-tuned.
Why is it different from other lineup tools?
Most all-time lineup tools just add up box-score averages, which rewards stacking five high-usage scorers. SweepSzn models finite possessions, usage overload, era-normalized stats, defense at full weight, spacing, and fit — and then explains, in plain English, what helped and what hurt.
Where does the player data come from?
Basketball-Reference box scores from 1950 to 2025, normalized per season so different eras can be compared fairly. Pre-1974 defense (before steals and blocks were tracked) is estimated honestly rather than guessed.
What are the game modes?
Eight ways to play. Daily gives everyone the same spins each day and ranks results on a verified leaderboard with streaks. Classic shows full stats while you draft; HoopIQ hides them so you draft from memory. Challenge sends a friend the same spins to beat your record. Then four modes that lean on the explainable engine: Factor Hunt asks you to predict which factor matters most before the reveal; Prime Draft drops the era reel so you can build cross-era fives with every legend at his peak; Blueprint has you commit to a tactical objective (spacing, defense, usage discipline, rim, or balanced) before the spin and grades your execution on that axis; and Surgeon diagnoses your lineup's single worst factor and gives you one targeted swap to fix it, scored on the win delta.
What makes Factor Hunt, Blueprint, and Surgeon possible?
They all depend on the engine explaining itself. Because every result decomposes into named factors — usage overload, spacing, interior size, perimeter defense — the game can ask you to predict them (Factor Hunt), grade you on one of them (Blueprint), or diagnose and fix your worst one (Surgeon). A tool that only outputs a number can't do any of that.