Data for a player prop model: what is available

Inputs by source

A prop model needs targets to compare against, player performance and context. This API has parts of each, from separate sources with separate identifiers. List what your model needs against this table before you write features.

What is not available

As of 2026-11-03, native sportsbook scopes cover New York NFL, NHL, college basketball and college football game lines and New York NHL Shots on Goal, points, assists, saves, goalscorers, power-play points and first-period shots. Read snapshots and streamed changes through the sportsbook endpoints. ESPN-attributed DraftKings moneyline, spread and total observations are also available across eight leagues. Current scope inventory is listed at /api/v1/sportsbook/scopes. That rules out features such as a sportsbook player-prop price or a consensus prop line from this API.

• Graded results. There is no public grading endpoint, so you derive any label yourself.
• Multi-season line history. Recording began around 2026-11-03. Retained history is short, and it is not a complete opening-to-close archive. It does not contain seasons of past lines, and there is no public graded-results endpoint.
• Injury reports, lineups and tracking data. These are not in the documented datasets.
• Canonical player IDs across all sources. See the join section below.

Reading Pick6 as an input

Pick6 rows carry standings multipliers such as 1.5x. A standings multiplier is not decimal odds, a probability or an established entry payout. That limits how you can use one as a feature. You can record it and test whether it helps a model, but do not relabel it as implied probability.

A player can have several rows for one stat at different targets, so pick a rule for the row you use and write it down. Skip rows where is_paused is true when you want an offer that could be selected. Skip stale rows for anything described as current. A target_value is a display target, not a forecast of the stat, so treat any such reading as your assumption.

python3 pick6-board.py --sport NHL --stat "Shots on Goal" --limit 25 --max-pages 2

Reading game logs

/api/v1/stats/player-games returns rows as {data, game}. data has athlete_id, name, team_id, opponent_id, category, did_not_play and a stats list. Each stat has key, label, display, numeric and scope.

curl --fail-with-body "https://draftkingsapi.com/api/v1/stats/player-games?sport=NFL&season=2026&limit=50" \
  -H "Authorization: Bearer $DRAFTKINGS_API_KEY"

• Use only scope: "game" values. A season-to-date value is not a game value and would leak later information.
• numeric is null for compound values such as 1-3 or 13:10. Parse display yourself if you need them.
• Filter on category client-side: boxscore for NBA, receiving, rushing or passing for NFL, forwards, defenses or goalies for NHL.
• Pages are ordered by cursor, not by date. Sort by game.starts_at before computing any rolling feature.

Identity joins and their limits

Each source has its own IDs. Pick6 entity_id is a DraftKings identifier and athlete_id in statistics is an ESPN identifier. There is no verified mapping between them, and the API will not infer one. In a 2026-11-03 sample of same-named NFL players, Pick6 entity_id matched DFS playerDkId for 128 of 129. Treat that as an observation to verify for your sport, not a guarantee.

A sportsbook event_id looks like espn-draftkings:espn:NBA:401909925, and the statistics event_id is the trailing 401909925. If you match players on name, normalize carefully, refuse ambiguous matches, record the method and report the unmatched count. See matching players across datasets.

A safe starting feature set

Start with what you can compute without leakage and without a shaky join. A baseline only needs one athlete's own earlier games. Add other inputs one at a time, once you have recorded them with timestamps.

1. Rolling mean of the target stat over the player's earlier games.
2. Minutes or role proxies where they exist as numeric game values.
3. Opponent and home or away from game.teams and opponent_id.
4. Recorded Pick6 target at a stored time, only after you have recorded it yourself.

The training data guide covers labels and splits, and the NBA model guide builds this baseline.

Check the recorded data

• Measured coverage report
• Sources and status

Related resources