Read Net Rating Before the Standings: Filtering Noise in the Regular Season

On the night of November 12, I stayed back in my Da Nang office with a spreadsheet open on screen. The first nine games of the team I track had just wrapped: 3 wins, 6 losses. On the other end of the line, the front office had already prepared two options — fire the head coach, or buy out the import. I switched to the second sheet and shared my screen.

The team's net rating after nine games: +1.9 points per 100 possessions. The standings said they had lost more than they had won. The underlying process said they were beating opponents by nearly two points every hundred trips down the floor. Those two facts do not contradict each other. They are measuring two different things.

The line went quiet for about four seconds. I am used to that silence. It usually shows up right before the question: "So why are they losing?"

Basketball is a sport where a single game carries very little information. Each team puts up 85 to 95 shot attempts per game, not counting the roughly one-fifth of possessions that end in turnovers or offensive fouls. The conversion rate on those attempts swings violently from night to night, and most of that swing has nothing to do with a team playing better or worse.

The number I use to describe this is simple. When the game count n is below 15, the standard error on net rating sits somewhere between 5 and 6 points per 100 possessions. That means a team at +1.9 after nine games could genuinely be a +8 team, or it could be a -3 team. My spreadsheet is not saying whether that team is good or bad. It is only saying the standings are not yet qualified to draw a conclusion.

For that reason, the first thing I open in the regular season is always net rating adjusted for opponent strength, not the win-loss column. After that come four supporting groups: opponents' three-point percentage against this team, clutch-time record, shot quality versus actual shooting efficiency, and free-throw rate. Those four share one trait — they are noisier than net rating, yet they are where most of the misleading narratives get born.

The regular season is a long race, and that very length turns it into an ideal environment for rushed conclusions. A team that wins four of five gets called a title contender. A team that loses four of five gets called a crisis. Both labels get pasted onto an information load that is very close to zero.

The standings measure the final result. Net rating measures the process that produced that result. A basketball game ends in two digits, and those two digits are heavily driven by the final possessions. Four of the nine games for the team I track finished within five points. They lost three of those four. If three late three-pointers had dropped instead of clanging off the rim, the standings would read 6-3 and nobody would have called me at all.

A team that wins seven games by fewer than five points is not stronger than a team that loses seven games by fewer than five points; it simply got lucky in the most expensive possessions. What worries me is that both the market and the team's own front office read clutch records as a mental quality, when the sample used to measure it is so small it borders on meaningless.

The second most misunderstood group of metrics is opponents' three-point percentage. Across a season, that number depends on who is doing the shooting. A defense that forces the ball into the hands of poor shooters will post a low mark. A defense that forces the ball into the hands of good shooters will post a high one. That means the metric contains both tactical signal and random noise, and the ratio between those two parts only stabilises after roughly 25 to 30 games. Before that mark, a perfectly respectable defense can rank 28th in the league on this number, simply because opponents made tough shots.

The third group is shot quality. I log shot location and the distance of the nearest defender, then convert that into expected points per shot. When the team I track shoots roughly four points per game below its expected mark early in the season, I do not propose changing the offensive structure. I propose holding still and waiting. Shot quality is a more durable variable than shooting efficiency, because it measures decisions rather than outcomes. Six weeks later, that team's shooting efficiency reverted to its expected level, and the record flipped without any tactical revolution.

The fourth group is schedule. Four of their first nine games came against top-tier opponents. Adjusted for opponent strength, that +1.9 net rating is actually +3.4. In the other direction, another team in the same league is sitting at 7-2 and getting praised everywhere; their adjusted net rating is only +0.6, because six of those nine games came against bottom-tier opponents. This is where data and perception split most cleanly. Fans remember faces. A spreadsheet remembers no faces, only numbers.

Every coach talks about feel. I have no feel; I have standard deviation. Which is not to say I dismiss professional intuition — the person in the coach's chair sees things the camera does not. But when a decision gets called genius only because it worked across a sample of seven possessions, I am obliged to open the log file and check again.

Based on my experience tracking games, the hardest part has never been the calculation. In 2026, while working with a club in the domestic league, I once recommended keeping the same rotation after a five-game losing streak. The coaching staff stayed patient for two weeks. The team won six of the next eight. I always remind myself that result does not prove my model right; it merely failed to disprove it. Those are two very different things.

Numbers do not lie, but they do not tell stories either. Hand a +1.9 net rating to someone who does not read stat sheets, and they will ask how many games the team won. Hand the standings to someone who watched none of the games, and they will conclude the team is weak. Both are missing half the picture, and that half lives where the two data sources meet.

Data is a monastery: the less noise there is, the more clearly you hear something trying to speak. The trouble with the regular season is that it keeps generating fresh noise — a shock win, an injury, a quote, a tweet. My job is to filter the noise without filtering out the signal along with it. People look at the box score to remember a game. I look at net rating per 100 possessions to understand the game that never happened.

Correlation is not causation, and this is where anyone holding a spreadsheet is most likely to shoot themselves in the foot. The biggest temptation is to call everything abnormal luck. Not quite. A defense can legitimately suppress opponents' three-point percentage by design: switching early on screens, forcing the ball into the middle, ignoring poor shooters. In that case, the low number is skill, not luck, and the regression to the mean that everyone is waiting for will never arrive.

My spreadsheet cannot see the locker room. It cannot see a player with a sore ankle still suiting up because the schedule is packed. It cannot see the commercial flights, the sponsor photo shoots, or the preseason friendlies a team is forced to play to recoup costs. Everything branded as load-management science tends to get romanticised, when in practice its main purpose is often to clear space for the off-court schedule. There is no column in my file that captures that.

Let me be explicit about the condition under which I am wrong: if that team, after 25 games, still holds a positive net rating but a winning percentage under 40%, then the problem sits exactly in the things I cannot measure. At that point the answer will not be in the log file. It will be in the meeting room, between people who can look at each other.

The next checkpoint I am watching is game 20 to 25, when the signal-to-noise ratio in net rating flips. By then, the standings will start saying the same sentence as the spreadsheet — or one of the two will be exposed as the liar. My team, at +1.9 net rating and 3-6, could finish the season third, or it could finish eleventh. The only certainty is that over the next fifteen games, someone will call me and ask the exact same old question.