Long-form investigations that start with a question and follow the numbers wherever they lead. No takes without a table behind them.
Winning teams ran on ~8 effective shooters, losing ones 12-15; r = -0.67 with wins. But diffusion tracks roster churn: concentration is a symptom, not a scheme.
Margins swell from 11.5 in autumn to 13.0 by spring; five-point games fall by a third. Split 1,230 games by tier and the creep is the bottom falling out.
Two of three spotlight benches inherited exactly .500 teams. Daigneault's completed arc, Redick's proof year, Dusty May's day zero, from real season tables.
Rank 547 shooters by load and the price of a shot barely moves: harder diets and better shotmaking cancel to 1.09 — until the last rung, where it breaks by 0.09.
Across 1,230 games, favorites separated by 5+ points of season win% won 70.8% — and 30-point mismatches still lost one in six. The curve and the upset ledger.
248 threes from 30+ feet hit 20.2% — a blend of a 28.9% logo shot and a 6.2% surrender, split by a dead zone at 33 feet where the arc simply ends.
The restricted area shoots 65.8%, the rest of the paint 44.2% — a 21.6-point cliff, five times any gap at the arc, and the fall is finished five feet from the rim.
Split 7,255 above-the-break threes by zone: wings 36.5%, top of the key 34.6% — the worst three on the floor. Match distances and the gap thins to 0.2 SE of noise.
Combined scoring climbed from 223.0 in October to a 234.4 December peak, then fell to 222.6 by March. The season's arc is a hump — with an 8-point post-break cliff.
Nearly half of Boston's 64 wins were 15-point routs; Charlotte managed one all season. Win style recovers half the quality signal a W-L record throws away.
Blown-out teams came back 21 points better in the rematch. Against a baseline that stops remembering the blowout, the revenge premium is +1.3 points of noise.
Replay all 1,230 games in date order: the East's top eight by record froze on December 16; the West's last chair changed hands until April 12, two days out.
All 435 season series of 2023-24, audited: sweeps ran at 1.9x the coin-flip rate, yet a .100+ gap in final W% took the series outright just 68.4% of the time.
Teams on a 3-in-4 won 43.8% of 2023-24 games against normal rest, but stacking the third game on a back-to-back moved the margin 0.05 points. Fatigue doesn't stack.
Corner threes hit 38.6% and pay 1.16 points per shot against 1.08 above the break. Fit accuracy to distance and the premium vanishes into geometry: 22 feet, priced.
Sided attempts split 50.3/49.7 and both sides shoot an identical 38.2%. The shooter's-corner myth, the one borderline signal, and why the floor has no grain.
Raw next-game win rates climb from 41% to 62% by previous margin — pure momentum, until you adjust for team quality and the effect collapses to r = +0.017.
Across all 1,231 games of 2023-24, the correlation between total points and final margin is −0.02. Scoring volume simply doesn't predict competitiveness.
Measure each team's home win% minus road win% in 2023-24 and the spread runs from the Rockets at +32 to the Grizzlies at −22, around a +8.6 league average.
Each point of net rating was worth 2.33 wins across 2023-24's 30 teams (r²=0.95), with the line through exactly 41 wins at zero. The margin is the standings.
The steadiest team of 2023-24 was Detroit (margin SD 12.0) — one of the worst; the wildest was Dallas (SD 18.0), which reached the Finals. Steady isn't good.
Effective FG% jumped from 47.1% to 54.7% over 2004-2024 while raw accuracy crept from 43.9% to 47.4%. The efficiency boom came from the shot mix, not the aim.
Compute each 2023-24 team's schedule from who it played and the spread is 1.3 net-rating points a game — with weak West teams facing the hardest slates.
Knowing exactly how good both teams are explains only ~26% of the variation in 2023-24 final margins. A game is a quarter signal, three-quarters noise.
A tired team facing a rested one won just 41.9% of 2023-24 games (margin -2.8), but two tired teams are back to 50/50. Rest is a differential, not a state.
The better team by net rating won 67.9% of 2023-24 games — a near coin-flip (55%) in toss-ups, up to 81% in mismatches. It all hangs on the talent gap.
Rebuild every 2023-24 win/loss sequence and compare it to weighted coin flips — the real streaks land almost exactly on the random null. Momentum isn't there.
In 2023-24, offensive rating predicted wins better than defensive rating across all 30 teams (r 0.87 vs 0.75) — and net rating (r 0.98) settles the slogan.
The most common NBA game of 2023-24 wasn't a nail-biter — it was a 20-point blowout. The real distribution of 1,231 final margins, and what it does say.
Drop 25,000 real shots onto the floor and 69% come from just two places — the rim and the arc — while the mid-range has collapsed to one shot in nine.
Across 1,231 real 2023-24 games, home teams won 54% and outscored visitors by 2.2 points a night — an edge that hides almost entirely in the blowouts.
Teams obsess over guarding the arc, but opponent three-point percentage barely repeats. What defenses can control about the three, and what's just variance.
Stats keep counting after the game is decided. How garbage-time minutes inflate numbers and warp plus-minus — and how analysts filter them out.
The box score stops at the player. Lineup data asks how five-man units actually perform — and why most lineup numbers are noisier than they look.
The same shots that win a title can lose a playoff series. Why a three-heavy league is a higher-variance league, and what that means for reading any single game.
The NBA defines clutch time precisely: last five minutes, margin within five. So I pulled it — and tested whether “clutch” is a real, stable skill.
Does a night of rest actually move the needle? I split a full season of games by days of rest to measure the real cost of the second night.