How the score works

The idea

A game can be worth watching for different reasons. Two top teams meet. A superstar is in form. Someone puts up a night people still talk about years later. Or it was simply close to the final second. Those reasons are independent of each other, so they are rated separately.

Each of the five categories yields 0 to 100. The total is their weighted mean. You set the weights yourself — turn down what bores you and the category drops out of both numerator and denominator, so it does not distort the rest.

The total is computed in your browser, not on the server. That is why the sliders act without a reload, and why every reader gets the same cached response no matter how they weight it.

1 · Team quality

How strong are the two teams? Per team: 50% win percentage, 30% the last ten games, 20% point differential per game, where a differential of 12 points per game counts as the maximum.

The two team values are not simply averaged. The mean counts 60%, the weaker of the two 40%. A top team against the bottom of the table should not look as good as two strong sides.

In the first ten games of a season, win percentage is noise — after three games a team sits at either 100% or 0%. Until then the rating leans on a prior. We have no previous-season strength in the database yet, so that prior is the league average. Which means: early in a season this category sits near 50 for every game and separates little.

2 · Stars

Not how much star power is on the roster, but how much of it actually played. A resting superstar does not make a game worth watching.

Every rated player has a tier: 1 for superstars (factor 1.0), 2 for All-Star level (0.6), 3 for familiar names (0.3). From 28 minutes played a player counts fully, below that proportionally.

The denominator is the best six players across both rosters. So the category answers: how much of the available star power was actually on the floor? An important side effect: when everyone is fit the value sits near 100 — the category separates games mainly when someone is missing.

The tier list is maintained by hand and is currently an automatic proposal from a review of last season, still unchecked. How many entries have been reviewed is reported at /api/health.

3 · Players from your nation

A separate value is computed for every nation with at least one player in the game. Which one you see is up to you in the settings.

A player counts from 10 minutes, fully from 24. Each weighted player is worth 40 points. On top of that come up to 30 points if that nation’s best player had a strong game — measured by game score, from 8 (no bonus) to 28 (full bonus).

A nation is listed even when its value is 0. Otherwise "nobody played" could not be told apart from "nobody played long enough".

4 · Standout performances

Did anyone do something extraordinary? The basis is the Hollinger game score of the game’s best individual performance. A game score of 15 yields 0, 25 yields 40, 35 yields 75, 45 yields 100 — linear in between.

On top come bonuses for the marks people remember: triple-double 10, quadruple-double 40, 40 points 8, 50 points 18, 60 points 30, 20 assists 12, 20 rebounds 10, eight threes 8, career high 6. The total stays capped at 100.

5 · Tension and flow

Was it close? This category comes from the course of the game: final margin, lead changes, time spent within five points, the last five minutes, overtime, comeback size and the win-probability curve. A fifth of it is how well the game was played — shooting, turnovers, pace.

A game that ends 25 or more points apart without a comeback is a blowout. Then this category stops at 25 — and so does the total. Without that second rule, star power alone lifts a game decided early above a close one; we observed exactly that across 200 games.

This category is the reason the spoiler protection exists. It gives away how the game ended.

Spoiler protection

The site should help you choose without ruining the game. So the split is not a display option but built into the mechanics.

Visible right away: team quality, stars, nation and standout performances, plus a total computed without the tension. Only after a click come tension and flow, the final score and the highlights. They are two separate switches, because each gives away something different.

The endpoint the page fetches on load does not know these values at all. They sit behind a separate request that happens only after your click. A test checks on every build that the HTML we ship contains no team, no final score and no tension value.

One exception you should know about: the blowout warning gives away that a game was decided early. It can be switched off.

Preview

For the coming night there are only three categories: team quality from the current table, expected stars from the injury report, and expected players from your nation. Tension and standout performances require a game to have been played.

Expected values are systematically lower than the ones afterwards, because the bonus for strong performances only exists later. "71 expected" can become 85 by the next morning. That is not a contradiction but the same number at two points in time.

If a roster is not known yet — early in a season not every team has played — the stars show a dash instead of a zero. A zero would claim there are no stars there.

Where the data comes from

From publicly reachable ESPN endpoints, with cdn.nba.com as a fallback. Both are unofficial and can change; every response is validated against a schema before processing, so a change shows up instead of quietly producing wrong numbers.

A night’s games arrive over several hours. The footer says when data was last fetched. If something is still missing, the page says so.

What the score cannot do

It measures neither sporting quality nor fun. It measures features we assume correlate with entertainment. Two top teams feeling each other out get high team quality and still low tension.

The constants on this page are calibrated, not proven. They come from a review of roughly 200 games of last season, with top and bottom lists checked by hand. If a score feels wrong to you, it is probably not an arithmetic error but a weighting worth arguing about.