The Clock · Part 3
The Clock, Part 3: Who gets there from here
How much of a deep playoff run is luck, and what the teams that made one actually did.
I started this one on Instagram. An edit of Tyrese Haliburton's Achilles came up, and I had the thought I've had since that night: man, it would have been cool if they'd won. So I looked up what the Pacers' title odds had been before that season. Plus 6600. The price says one in 67. How does a team go from there to Game 7 of the Finals against the best team in the league, looking like they might take it? And how much of it was just dumb luck?
That's the question, and this piece points it at Minnesota. Part 2 left the model at 1.79% to win the title, with a 7th seed as the most likely April. That's a long way from a parade, and it's not nothing. So this one is about the ground in between: from where this team starts, how do teams get as far as a conference final, and how much of getting there is up to them? A conference final rather than a title because that's where there are enough teams to learn from. The Finals are the same road, further on.
So I went back through every playoff series since 1997-98, all 435 of them, and asked each one the same two questions. Given how good the two teams were, what should have happened? And when a team got more than that, where did the extra come from?
Two springs, one coach
Start with the answer to the Pacers question, because it's the one I came for. The run to the conference final, which is the part I can measure, was about four-fifths luck, and I'm fairly sure of that number. Their path should have gotten them there about one time in ten. With everything that broke right, it was two in three. And it wasn't the injuries. Milwaukee lost Damian Lillard to a torn Achilles in Game 4 of the first round, and Cleveland was without Darius Garland for two games and Evan Mobley for one, but by my measure health on both sides is a small slice, and the draw actually worked against them. The luck was shooting. The biggest piece was Cleveland going cold in the second round, and the next was Indiana's own shooting in both series. If you're a Pacers fan, so what. Nobody hands the conference trophy back.
Here's the one that got me, though. I ran the same math on the 2011 Mavericks. Dirk's title. The run everybody remembers as a seven-footer refusing to lose. Before I ran anything I wrote down "earned." The numbers came back four-fifths luck. Same as the Pacers.
Two of the luckiest deep runs in the whole sample, seventh and eighth of the 116, and the same man was on the bench for both of them. Rick Carlisle. Hold onto that, because it's either the best argument for luck in this whole piece or the best argument against my method, and I'm not sure which.
Each piece is its part of how unlikely the run was, over the two series that got the team to a conference final. Own shooting and the other team's shooting are the threes that fell or missed beyond what the shot profiles explain. Health is both rosters against their own norms. The draw is the bracket: below zero, it made the path harder, so it takes that much back. The marker is the run's total luck share, with its 80% band from resampling; a share of 0.5 or more is mostly luck. The methods page has the split. Hover a piece.
Most of the Mavericks' luck is one series, and you know the one. Second round, the two-time defending champion Lakers, and Dallas sweeps them. Dallas made 49 of 106 from three in four games, way over anything its own season said to expect. At the other end the champs went cold: 15 of 76. Put together, that's close to ten points a game of Dallas shooting over its head and about eight more of the Lakers shooting under theirs. In a series Dallas won by fourteen points a game, that's the series.
And most of Dallas's part is one afternoon. Game 4, in Dallas, on Mother's Day. The Mavericks made 20 of 32 from three, which tied the playoff record at the time. Jason Terry took ten and made nine. Peja Stojakovic took six and made six. The champs came apart in the fourth quarter: Lamar Odom was thrown out for a hard foul on Dirk, and Andrew Bynum followed him less than a minute later for a flagrant foul on J.J. Barea. That one game holds about seven in ten of Dallas's hot shooting for the series. The Lakers had already had their worst night back in Game 2, when they made 2 of 20.
Each game has two bars, Dallas on top. The light track is threes taken, the filled part threes made, and the tick what the team's own regular season, adjusted for the other team's defense, said to expect from those attempts. Dallas's expected rate was .341 and the Lakers' .337. Dallas made 49 of 106 for the series and the Lakers 15 of 76. Hover a bar.
So was that luck? Here's where I have to be honest about what the math can and can't see. There's no tracking data before 2013-14, so I can't tell you how open those shots were. What I can tell you is that 48 of Dallas's 49 made threes came off a pass. That looks less like a team getting hot and more like a team moving the ball until a champion's defense gave up the shot it wanted, over and over, for four games. Kobe Bryant was playing on an ankle he'd turned again in the first round, and it turned out to be Phil Jackson's last series as a coach. My method is built to catch a team shooting over its head. It can't tell a hot week from a game plan that broke an old, hurt champion, and I'm not going to pretend it can. When Carlisle's name shows up twice among the luckiest runs, that's the method telling on itself a little.
Keep that in your pocket. It comes back when we get to April.
What luck usually is
How I scored it, in one breath. Take any playoff series. The two teams' regular seasons and who has home court tell you roughly what should happen. Then look at what did happen and ask where the gap came from. I was willing to call three things luck, because none of them carries from one year to the next: a team shooting over its head for two weeks, a team being healthier than it usually is, and the draw. A hot April tells you nothing about the next one. I checked.
Not every hot streak is luck, either. Some of it is a team choosing to play differently in April: more threes, more trips to the line, one more pass before the shot. I pulled that part out and called it shape, because it's the part a coach actually draws up. It's real and it's tiny. It explains about one percent of the shooting swings in a series, and only one run in the whole group got as much as a point a game out of it. I had to fix my own rule to get there, and the methods page shows the before and after. Shape tells you what a run looked like. I wouldn't use it to plan one.
So the Pacers and the Mavericks are the exception. For the typical team that made a conference final, luck covered about a sixth of the distance. For the 45 teams that started where the Wolves are starting now, about a fifth. Give the defense credit for the part of the opponents' cold shooting that kept showing up round after round, which is fair, and both drop to a little over a tenth.
Put that in teams. Of the 116 that made a conference final in this stretch, 28 got there mostly on luck and 88 mostly on being good. Forty of them sit close enough to the line that I wouldn't swear either way. Strip those out and it's 14 clear luck runs against 62 clear earned ones. Good beats lucky, and it isn't close.
And the bracket, the thing every fan blames first? For the typical run it rounds to zero. The draw didn't make these runs.
Minnesota has one at each end. The 2024 team, the one that beat Denver, got to the conference final against its luck, and I'd stand behind that one. The 2004 team, Garnett's, leans a little toward luck, and that one could go either way. The 2019 Raptors, Kawhi's one year there, earned theirs too, probably against their luck. The 2023 Heat, Jimmy Butler's play-in team that made the Finals, are the one the method can't settle: about seven-tenths luck, close enough to the line that it could go either way.
Each row is one team that reached a conference final, sorted by luck's share of how unlikely its run was: shooting nobody could plan for, health and the draw. The bar is the 80% interval from resampling. Right of the dashed line, at 0.5, the run was mostly luck. Shaded rows are too close to call: their interval crosses the line. Below zero, the team got there against its luck, and a run past the axis sits on its edge. Hover a row.
Where the Wolves are standing
Now the board. Start with the teams that were standing where the Wolves are standing. Not the ones with the same names, the ones with the same record: a net rating anywhere from a couple of points under water to plus five, and a playoff exit in the first two rounds or no playoffs at all. Minnesota was +3.1 last year and went out in the second round. That's the group.
There have been 426 of those teams since 1998-99, and this is what the next season did to them. A bit more than half made the playoffs. A quarter won a round. One in ten made a conference final. Seventeen made the Finals. Seven won it all. Seven, out of 426. And it isn't an old-days thing: since the play-in came in, it's 9 of 93 reaching a conference final, the same one in ten.
Now the part I didn't expect. The model from Part 2 never looked at any of this. It builds the Wolves up from the roster, one player at a time, and plays out the season, and it lands in the same place: a round about 28% of the time, a conference final 10%, the Finals about 4%, the title 1.79%. History's ladder says 25, 11, 4 and 2. Two completely different ways of asking the question, and they agree. When that happens, you start to believe the number.
The band: every team whose net rating the season before was between -2.5 and +5 and that went out in the first two rounds or missed the playoffs. Each bar's width is the share of the band that got at least that far the next season. The green outline is the model's Minnesota from Part 2 (the frozen model, headline mix), drawn at the width of its chance on that rung; the model has no playoff-qualification figure on this basis, so that rung has none. Hover a bar.
And since Part 2 left you hanging on the champions: the last eleven were all top eleven in net rating twenty games in, call it Thanksgiving, and ten of them finished the regular season top five. That's the bar. The Wolves aren't there yet, which is exactly why this piece is about the teams that weren't either.
The cheat codes that don't work
So if luck is a fifth and the bracket is nothing, most of a deep run is the team being good. Fine. What did the teams that got good do to get there? I went looking for the cheat code: every decision a team makes before the playoffs that I could put a date on and measure, each one checked against teams of the same quality at the same point in the season, so a move only gets credit for what happened after it. The graveyard first.
The trade deadline. This is the one Wolves fans will be screaming about in February, so hear it now. Of the band's teams that made a conference final, a little more than a third added a rotation player at the deadline or off the buyout market. Of the band's teams that fell short, about half did. The teams that went deep made fewer of those moves, not more. I'm not saying the deadline hurts. I'm saying the history says it isn't where runs get made, and Tim Connelly's phone is not the lever.
Load management. Teams that rested their top eight more than the typical team weren't any healthier come April, as far as I can tell. And on the nights those guys sat, their teams played about a point and a half per hundred worse. You pay the tax in January and the refund never comes.
Changing coaches midseason. Teams tried it thirty-five times. They did a little worse afterward than teams like them, well inside what chance gives you. A shrug.
The top-ten player. The one everybody reaches for first. Once you know a team's net rating, knowing it has a star didn't make the forecast any better at any checkpoint. With this many seasons a real effect could still be hiding in there, but it didn't show up. And the uncomfortable part for us: by last season's box-score rating, Anthony Edwards was 27th in the league and LaMelo Ball 29th, so Minnesota wouldn't have qualified anyway. If you watched Ant last year and think that's a problem for the rating and not for Ant, you and I agree, and that's a different article.
One more that isn't a cheat code so much as a warning label. Teams whose top two creators were in their first season together played a little worse through the middle of the season, about half a point per hundred, and it leaned that way in both halves of the sample. It didn't clear the bar I set for calling something real; test fifteen things and one of them leaning by chance is about what you'd expect. But Edwards and Ball are a first-year pairing, so if the Wolves are under the lines in January, that's the first place I'd look.
The thing I thought worked
One thing came back alive, and for about a day I believed it.
When a coach changed his starting five with nobody hurt, and stuck with the new group for at least five games, his team played about a point per hundred better over the next twenty games than teams of the same quality that stood pat. It held in both halves of the sample. It was the only thing on the list a coach can just decide to do on a Tuesday, and I had the paragraph written, Chris Finch's name and all.
Then I went looking for examples to show you, and the examples told on the finding. Two of the biggest "lineup changes" in the pile were LeBron James coming back in January 2015 and Dirk Nowitzki coming back in January 2013. The rule I'd written down caught any change where the guy going to the bench wasn't hurt. It never asked whether the guy coming in had just been. Seven in ten of those fourteen hundred changes were a player returning from an injury or a spell out of the rotation. Another one in seven was a starter getting his old job back. Of course those teams got better. Their guy was back.
Strip all that out and ask the question I actually meant, a coach promoting a bench player who'd been available the whole time and sticking with it, and it has happened 67 times in the whole sample. Those teams leaned the same way, and I can't tell. The spread is too wide, and the two halves of the sample point in opposite directions. The clearest ones at either end: Jerry Sloan moving Antoine Carr into Utah's five for Greg Foster in December 1997, and the Jazz were about six points per hundred better over the next twenty games than teams like them. Frank Vogel moving Jonathon Simmons into Orlando's five for Terrence Ross in November 2017, and the Magic were about five and a half points worse. Most of the rest land in between, on both sides of zero.
So the one lever I thought I'd found is a mirage, at least as far as I can see it. I'm leaving the first version on the methods page next to the second, because that's the whole point of writing the rules down before you run anything.
In Minnesota, who starts and who plays the four is still Chris Finch's call. I just can't show you that any one of those calls, on its own, moves the number.
Each row is one lever tested before the playoffs: the dot is its effect and the line its 95% interval, against no effect at zero. The groups keep their own units. Rest is playoff health against the team's norm, in points a game; the in-season moves are net rating per 100 possessions against same-quality teams; the playoff habits are margin a game per standard deviation, measured before each series. The lineup lever shows twice: the solid dot is the clean version (a bench player promoted with nobody hurt), the hollow one the first version, before the fix. A lever counted only if it leaned the same way in both halves of the sample and survived the correction for testing 15 at once. The trade deadline and the top-ten player were tested differently (a share comparison and a forecast test), so they aren't on these scales. Hover a row.
And here's the sentence I'd put on the wall. At every checkpoint from game twenty to April, nothing I tried predicted a conference final better than net rating alone, on seasons the model had never seen. As boring and obvious as it sounds, being good is what matters. Almost everything else is noise.
What April looks like
This is where the Carlisle problem comes back. The method can't tell a plan from a hot week, so all I can do is show you what the deep runs looked like and let you decide. Each of these is measured in the same series it describes, so it can follow winning as easily as cause it. Teams that beat what their regular seasons said they'd do took more threes than they had all year, got to the free-throw line more, and won the close games. Two of those come with an asterisk: some of the free throws are the other team fouling late to stop the clock, and winning the close games is close to saying "won" twice.
Minnesota's 2026 playoffs went the other way on the first two. Against Denver and again against San Antonio, a smaller share of the Wolves' shots were threes than in the regular season, and they got to the line less often. I'm not writing a prescription off that. I'm putting it next to what the deep runs looked like and letting you look.
The lines
So here's the yardstick, the simplest thing that held up: take a team's net rating at a point in the season and read off how often teams like that, every team since 1998-99, went on to a conference final.
Before the season I drew two lines for this team: 14th in net rating by Thanksgiving, and 9th by April. In net rating, those are roughly +1 and +3. A team sitting right on the first line has gotten to a conference final 7.8% of the time, about one in thirteen. Make the April line and it's 12.5%, about one in eight. Get to +4 by April and it's about one in five. From there it climbs fast: a +6 team is at 44%, a little under a coin flip.
Each dot is every team-season from 1998-99 to 2025-26 whose net rating at the checkpoint fell in that one-point bin (the end bins hold everything past them); its height is the share that reached a conference final, its size the number of teams. The line is the yardstick, a logistic fit on net rating alone (nothing else passed its test), with its 80% band from resampling seasons. The dashed vertical line was set before the season: 14th in net rating after 20 games, 9th in April. The dashed level is Part 2's simulation of Minnesota's season. The key under the plot gives each reading in the article's words; hover a mark for the exact chance.
Part 2's simulation gives them 10% for a conference final, right between the two lines. Two different machines, built different ways, landing in the same neighborhood, and I'd trust the neighborhood more than either decimal.
The tracker
Once the season starts, the stats page tracks all of this every Monday and freezes it at each checkpoint: what the yardstick gives Minnesota at its current net rating, where that sits against my two lines and Part 4's, and the top eight's health against their own norm, as context.
It also logs any starting-five change of the kind I looked at, the day it happens: a bench player promoted with nobody hurt. Each one shows as provisional until there's enough season on both sides of it to compare, then what the team did over the next twenty games against teams like it. Context, not a verdict. The lines don't move. That's the point of drawing them now.
The way there
From here, a conference final is about one in eight if they're 9th by April, and about one in five if they're a +4 team. Nobody gets to plan on the Pacers' spring. The way there is boring, and it's theirs.
Corrections
No corrections logged.