Polymarket NBA data: game-night pricing at 250ms

Basketball on Polymarket spans NBA and WNBA games, quarter markets, moneylines, and totals. The live ladder is free from the official API; the historical ladder is not available from it at all.

Because /book has no date parameter, 'how was the fourth quarter priced on that close game?' is only answerable through an archive that recorded the depth as the game played. PolyOrderbooks captures full L2 ladders four times per second for the basketball series it lists.

Basketball leans on the catalog-wide shapes: over/under totals (372k), quarter markets (23k), and matchup props. Each contract trades as its own book and settles independently.

All of them share the archive record — full bid/ask ladders with sizes, a captured_at timestamp, and the winning outcome on resolution — so a fourth-quarter book is recoverable exactly as it stood.

• Live ladder — official /book, free and keyless, current moment only (what the API serves).
• Historical depth — full L2 books at 250ms over REST, free Starter with 3 days, paid windows to 120 days.
• Discovery — /v1/series for NBA and WNBA; /v1/markets for questions and slugs.
• Coverage — measured counts per sport on the sports data page.

A prime-time NBA game: the moneyline sits stable through the first half, a 12-0 run reprises the winner inside one timeout, and the fourth quarter settles it with real market traffic. The official API shows those frames live; the archive keeps the fourth-quarter window so the run can be replayed exactly as priced.

Basketball trades across NBA and WNBA games, quarter markets, moneylines, and totals inside the 700k+ long tail. Quarter binaries settle on their own clock, so a single night generates a sequence of closed books — first quarter, second, third, fourth — each querying like its own fixture.

Three documented dynamics fit the record: whether fourth-quarter teams trade systematically tighter than first-quarter lines, whether a double-digit third-quarter lead collapses the game line to a few cents under the bound, and whether garbage time produces wide final-total quotes after the benches empty.

The standard pull applies: series and market endpoints scope the night, and the archive returns 250ms ladders for any quarter or the full game window — free to replay on Starter.

The quarter boundaries decompose the night into four independent books, and the minute after a timeout following a big run is where the executable side changes.

As long-tail data, single games are case studies; season-long windows belong in the archive for real estimation.

3 days free holds several nights; Pro stretches to 120 and resolved games remain queryable.

Study one: measure whether fourth-quarter books trade tighter than first-quarter lines, splitting by one-sided share so the executable side is the comparator on every frame.

Study two: replay a 12-0 third-quarter run and count how many seconds the game line spends collapsing toward the bound after the timeout; the one-second pattern from the crypto corpus is the reference rate.

Study three: test garbage-time totals drift once benches empty — whether the final-total quote widens by more than a band in the last two minutes.

A single night of NBA at 250ms fits the free window; a full week in Pro's 120 days gives the study enough games to separate momentum from noise.

First exercise: one quarter. Pull the second-quarter window of an NBA game from the series list and count the executable-side flips during a momentum run. The timeout marker is the natural boundary, and the archive keeps the whole run replayable after the buzzer.

Then pull the game moneyline at the same second and compare how much wider the quarter book trades; the multi-book structure of the night becomes visible in that single juxtaposition.

Resolve the quarter and game markets and keep them queryable; one week of nights is the honest first dataset, and the Pro window stretches that to a full month of games.

• Sports data
• Market data
• Order book data