NBA betting analysis — every market, explained

By Johnathan Cruzen-Myer · How this is measured

Every NBA page on this site, and what each is for. The player box scores landed in October 2026, so the prop markets are built on 56,571 appearances rather than on nothing — and the prop page leads with what survives a shuffled null rather than with a strike rate.

Tonight’s slate Every game on tonight, with the line, the total and how each side has been scoring. 10 games on tonight, read against 5,320 completed games. Player prop trends★ Points, rebounds, assists, threes and the combined lines — each split measured against that player’s own rate, and tested against a shuffled null. 121,146 graded selections at 73.5% against an own-rate baseline of 72.8% — an edge of +0.8 that 200 shuffles never reached. Model tuner Tune the walk-forward model for win/loss, the spread and the total, backtest it, and save each market on its own. Best picks Up to three player props per game, where his last 20 games beat the posted price and the expected return stays positive. Half markets First and second half lines and totals, priced off the share of the game each half actually carries. Quarter markets The same for each quarter. The benchmark is the measured share, not a quarter of the game total. A first quarter carries 25.2% of the game's points and a fourth only 24.2% — so an even split is the wrong benchmark. Trends this week Situational trends live on this week’s games, each shown against the matched baseline it has to beat. Team rule lab Build a rule from venue, rest, favourite status and form, and see it tested against a shuffled null. 5,252 games searched for a rule that beats the market. None survived the null. Streaks Runs of covers, overs and wins, with the probability each one is simply luck. 14 of 36 streak buckets beat chance, against 1.8 expected — the strongest signal of the four codes. Every club One page per club: splits by venue, rest, favourite status and period. One page per club, 30 of them, built from that club's own games. Best bets The highest-rated selection across all four codes each round, with the full walk-forward record behind it. 4 of 26 backtested cells turned a profit, and 0 did so beyond chance. Break-even is 52.4%.

Informational only, not betting advice · 18+ · Gamble responsibly. Help is available at Gambling Help Online.

How the NBA pages here are built

Every NBA page on this site is built the same way as the NFL ones: a claim is only worth printing if it carries the number of games behind it and the baseline it has to beat. For a period market that baseline is the share of the game that period actually carries, measured, rather than a quarter of the game total — an assumption that made first quarters look like a 31.5% under and second quarters a 63.9% over when neither was true.

That gap is now closed. The player game logs were ingested in October 2026 — 75,245 player-games across two seasons — so the player prop page is built on 56,571 countable appearances rather than on nothing. Preseason and All-Star fixtures are excluded from it, because thirty minutes in October against non-NBA opposition is not evidence about a January scoring line.

One NBA finding is worth naming because it runs against everything else on this site. Across 36 streak buckets per code, the NBA is the only code where the number of buckets beating chance is well clear of what chance produces. That is not a betting recommendation — it is a result, with its arithmetic shown on the streaks page, and it should be read alongside the fact that the book sees the same streaks.

Frequently asked questions

Do the NBA prop trends beat anything?

Across every selection they beat each player's own rate in the same market by 0.8 points, and 200 shuffles of the outcomes within each player's own games never reached that. The edge is real but small, and no single market can be singled out: under the same null the best of eighteen markets averages 2.25 points, which is about what the best one actually did. There are no posted NBA prop prices on file, so none of it can be turned into a return.

How are quarter and half markets priced?

From the measured share of the game total each period actually carries, taken as the median of period points divided by game points, rather than assuming each quarter is a quarter of the game.

Do you publish results when the model is wrong?

Yes. Every backtest on the site reports the losing cells alongside the winning ones, and the count of tests run, so the expected number of false positives is visible.