Is a "kill" anything more than a good stop rate?
Coaches and broadcasters call three defensive stops in a row a "kill," but no league keeps it as a stat. This study counts every kill in 31,197 NBA games from 2000-01 to 2025-26 and 3,880 WNBA games from 2009 to 2026, 6,543,543 possessions in all, and asks whether a kill measures anything a defense's stop rate does not. The short answer: it does not. Kill rate follows stop rate almost exactly, stops arrive in no streakier an order than chance would give, and the best kill defenses are the best stop defenses. The counting does leave a record book worth keeping.
Counting a stat that nobody keeps
A stop is an opponent possession that scores no points. Possessions come from a public archive of NBA and WNBA play-by-play that is already split into possessions (built with the open-source pbpstats library), so an offensive rebound continues a possession rather than starting a new one, and points come from the league's play-by-play score timeline. Every three straight stops is a kill: six in a row are two kills, and five are one. A streak carries across quarters and ends at the final buzzer. These rules were set on October 19, 2026, before anything was counted, so the results could not shape the definition.
That archive stops at NBA 2024-25 and WNBA 2025. The two newest seasons come from a second possession builder, corrected two ways and checked against the archive on four seasons both sources cover (the last section below). Possession points capture at least 99.85% of every season's points; the rest are free throws that belong to no possession. Every table was computed in Python and again in R, and the two agree on every cell.
A kill rate is a stop rate in different units
Across team-seasons, the correlation between stop rate and kill rate is 0.948 in the NBA and 0.929 in the WNBA, so stop rate alone explains 90% and 86% of the variation in kill rate. The WNBA's higher kill rate (7.93 per 100 opponent possessions, against 6.87 in the NBA) comes with a higher stop rate (52.5% against 49.6%).
Stops do not cluster
The sharper test holds each game's stops fixed and asks whether they bunch together. If every ordering of a game's own stops were equally likely, how many kills should a team have had? An exact calculation answers that, and because it keeps each game's stop count, it already absorbs opponent strength, pace, and blowouts.
The chart below runs that calculation for the NBA's best kill defense, and it shows why the order of stops, not their number, is the test: Milwaukee's kill total sits inside what its own stops give in random order, far above what an average defense would get from the same possessions.
| League | Team-seasons | Kills | Excess over random order | z | Team-seasons with abs(z) > 1.96 |
|---|---|---|---|---|---|
| NBA | 776 | 407,654 | -661 (-0.16%) | -1.85 | 5.5% |
| WNBA | 221 | 48,310 | +96 (+0.20%) | +0.83 | 4.5% |
About 5% of team-seasons would pass 1.96 by chance alone, and 5.5% and 4.5% do. They split evenly by direction: in the NBA, 21 team-seasons sit above +1.96 and 22 below -1.96; in the WNBA, 5 and 5. So no defense strings its stops together more than its stop rate implies. The WNBA's pooled count sits at chance. The NBA's is slightly below it, with 18 of 26 seasons under zero: stops alternate with scores a little more often than random order would give, by a fraction of a percent. The cause is not tested here.
A first version of this test gave each possession its game's stop rate as an independent chance, and it made every NBA season look anti-streaky. That shortcut overstates how often three stops line up, so the study uses the exact random-order baseline instead.
The best kill defenses are the best stop defenses
The top NBA kill rate is 2019-20 Milwaukee at 8.60 per 100, from the best stop rate of any NBA team-season here (54.6%), and its own streakiness z is -0.73. In the WNBA, 2021 Connecticut leads at 9.88 with the league's best stop rate (57.2%) and a streakiness z of +0.73. The 2003-04 NBA season holds four of the top seven spots (Detroit, San Antonio, New Jersey, and Indiana).
The Milwaukee comparison above holds across the NBA's top ten: each kill rate sits close to what its own stops give in random order, and the distance from the league average comes from stopping more possessions.
Of the NBA's top ten, one team-season clears 1.96 on streakiness: 2000-01 San Antonio, at 8.43 kills per 100 and a z of +2.66. The highest of any team-season is 2012 Los Angeles in the WNBA at +3.03, and the lowest is 2018-19 Dallas in the NBA at -3.79.
Twenty-two straight possessions without a point
On March 1, 2006, the Clippers held New Orleans/Oklahoma City scoreless for 22 straight possessions, from 7:45 of the third quarter to 7:50 of the fourth, starting from 7 points down. Cleveland matched it against Milwaukee on December 6, 2009, from 4:13 of the first quarter to 5:42 of the second, also from 7 down. The WNBA record is 21, also reached twice: Minnesota against Indiana on August 18, 2017, and Indiana against Seattle on June 22, 2023.
A second possession source, checked before it was used
NBA 2025-26 and WNBA 2026 come from live-feed possession builds. As stored, those builds hold empty end-of-period stubs and give period-start possessions to the wrong team, which would add 0.35 to 0.43 kills per 100 to every team on the overlap seasons, a false jump in the newest seasons. Corrected, they reproduce the archive's team kill rates with a correlation of at least 0.995 and mean team differences of -0.018 to +0.029 per 100. For scale, the spread across teams is 0.47 to 0.97 per 100.
Against that, the NBA's kill rate went from 6.75 in 2024-25 to 6.51 in 2025-26, and the WNBA's from 8.01 in 2025 to 7.33 in 2026, as points per 100 opponent possessions rose from 114.8 to 116.2 and from 104.5 to 108.9. Both drops are many times the mean builder difference.
What this does not show
The random-order baseline treats a game's possessions as interchangeable. Possessions late in close games or in garbage time are not, which can push the observed count either way, and the study does not model that. Every possession counts, including end-of-quarter heaves, as a broadcast would count them; the study has not measured whether heaves add kills evenly across teams. Free throws that fall outside any possession, such as technicals, are not charged to the defense. Record-book margins come from summed possession points, so they can differ from the true score by an unassigned free throw. Twelve games that pass the score checks are missing from the possession sources. And the newest seasons come from a different builder: a change of a few hundredths of a kill per 100 into them could be the builder rather than the game.
Method, code, and verification
The full technical write-up (data provenance, model specification, validation gates, and stated limitations) is the README. The analysis is implemented independently in R and Python and reconciled to numeric tolerance; run_checks.sh replays every gate.