NBA Shot Chart & Shot Location Scraper

Every NBA field-goal attempt, with where on the floor it came from — and the spatial fields the league does not publish, computed and verified against a full season.

No API key, no login, no browser. Filter by game, player, team, date or season.

What you get

One row per shot. Real output from a 2026-04-12 game:

Date	Player	Team	Made	Zone	Our distance	NBA's	Angle	Expected pts	Pts over expected
2026-04-12	Luka Garza	BOS	✅	Above the Break 3	25.778	25	28.75°	1.038	+1.962
2026-04-12	Franz Wagner	ORL	✅	Above the Break 3	27.777	27	27.67°	1.038	+1.962
2026-04-12	Paolo Banchero	ORL	✅	Restricted Area	1.985	1	40.91°	1.342	+0.658
2026-04-12	Paolo Banchero	ORL	❌	Restricted Area	2.766	2	−77.47°	1.342	−1.342
2026-04-12	Desmond Bane	ORL	✅	Mid-Range	18.623	18	35.82°	0.800	+1.200

Each row also carries the raw LOC_X/LOC_Y, plot-ready coordinates for the standard 500×470 court grid in both orientations, the game clock as elapsed seconds, home/away and opponent, and NBA's three zone labels.

What makes this different

The distance NBA publishes is wrong by design. SHOT_DISTANCE is floored, not rounded — verified across all 219,160 shots of the 2025-26 regular season, where floor matched every row and round matched 54%. Every published distance therefore runs short by up to a foot, always in the same direction. You get the exact value and NBA's integer beside it.

Expected points, from NBA's own numbers. The endpoint ships a league FG% table per zone alongside every query. Joining it turns a made/missed flag into points over expectation — shot-making separated from shot selection, with no model to trust. The baseline is fetched once for the whole season, deliberately: scoring a single day's shots against that same day's average drives the number to zero by construction.

A season query that is actually complete. shotchartdetail truncates at 102,400 rows and reports nothing — a 2023-24 season request returns a valid 200 whose data stops on 17 January. Summed month by month the same season holds 219,527 shots, so 53% is dropped silently. This Actor splits every league-wide query into windows before asking, and re-splits anything that still comes back at the cap.

Home/away that is right. The obvious source for it, MATCHUP, is usually written from the row team's point of view — in one captured slate, 17 rows of 18 were, and Orlando's own row read MEM @ ORL. Deriving the side from that string mislabels roughly one row in twenty. This uses the shot row's own home-team field instead.

Works in the offseason. The default run needs no input and returns the most recent day actually played, resolving back a season when the newest one has been announced but not started.

Modes

Give it	You get
nothing	The most recent day of play, league-wide
gameIds	Every shot in those games (~180 each) — the cheapest way to run it
playerIds	That player's whole season (~1,300 shots)
playerIds + gameIds	Just those players, in just those games
teamIds	That team's season (~7,400 shots)
dateFrom / dateTo	League-wide over that window, split automatically
season + seasonType	Regular season, playoffs, pre-season, play-in or All-Star

maxItems caps both the result count and the cost. IDs are NBA Stats IDs — 201939 is Stephen Curry, 1610612744 is Golden State.

Why this is cheap to run

Runs in 256 MB with no browser and no proxy on a healthy run, so the apify-actor-start fee is charged once rather than four times. A default run finishes in 3.5 seconds; peak memory plateaus around 160 MB no matter how long the run is.

What this Actor does not return

• Anything that is not a field-goal attempt. Made and missed shots only — no free throws, rebounds, assists or turnovers.
• Modelled shot quality. expectedPoints is the league's average from that zone. It is an honest baseline, not an xG-style model — defender distance and shot clock are not in this feed, so no number here pretends to include them.
• Player tracking or defender data. Served by different stats.nba.com endpoints with a different shape; not joined here.
• Seasons before 1996-97. Shot locations do not exist before then, and the Actor refuses those rather than returning an empty result that looks like a quiet day.
• SHOT_ATTEMPTED_FLAG / GRID_TYPE. Constant on every row, so emitting them would cost you storage and tell you nothing.

Example runs

{ "gameIds": ["0022500578"], "maxItems": 200 }

{ "playerIds": [201939], "season": "2025-26" }

{ "teamIds": [1610612744], "season": "2025-26", "seasonType": "Playoffs" }

{ "dateFrom": "2026-01-15", "dateTo": "2026-01-31", "maxItems": 20000 }

Notes

Unofficial and not affiliated with or endorsed by the NBA. Data comes from the same public endpoints nba.com's own statistics pages read.

docs/architecture.md records every measurement behind the claims above, including the header/TLS matrix, the cap proof and the full-season geometry verification.