CTG Pro Data Feed

Appearance

Changelog ​

Upcoming ​

These changes are released and will appear in newly processed games, but not yet backfilled for all games.

Added ​

  • New is_quick_no_dribble_shot field on Shots data set - indicates whether a shot came after a short touch with no dribbles, serving as the current catch-and-shoot equivalent.
  • New led_to_event fields on Ball Screens, Off Ball Screens, Advantage Creation 1v1 data sets - added led_to_event, event_type, pbp_event_id, and event_frame_number_he to identify when any of these actions is deemed to have led to the chance ending event.

Fixed ​

  • Shot end frame calculation on Shots data set - fixed rare cases where a shot end frame could be missing when ball tracking disappears after release before the normal shot-end trajectory logic can identify an end frame.
  • Duplicate ball screen entries on Ball Screens data set - fixed edge cases where one ball screen action could incorrectly appear as multiple consecutive entries.

[0.4.0-beta] - 2026-05-01 ​

Added ​

  • New ballhandler_movement_side field on Ball Screens and Drives — indicates whether the ballhandler moved left or right during the action when that context is available.
  • New ballhandler possession-touch linkage fields on Ball Screens, Off Ball Screens, Advantage Creation 1v1, Closeouts, and Drives - added ballhandler_outcome, ballhandler_possession_touch_id, and ballhandler_possession_touch_outcome_id where applicable, plus drive_id links on ball screens, off-ball screens, 1v1s, closeouts, and Possession Touches.
  • New shot make probability fields on Shots data set - added league_avg_make_probability_at_release and player_adj_make_probability_at_release.
  • New pre-shot touch fields on Shots data set - Added num_dribbles_during_touch (number of dribbles during the touch that preceded the shot) and touch_time_before_shot (duration of that touch in seconds)
  • New shooter movement context fields on Shots data set - Added shooter_velocity_before_shot, shooter_velocity_after_shot, shooter_acceleration, shooter_rim_movement_angle_before_shot, shooter_rim_movement_angle_after_shot, and num_close_teammates
  • New is_direct field on Drives, Closeouts, Passes, Ball Screens, Off Ball Screens, and Advantage Creation 1v1 — indicates whether the action ended in a direct scoring outcome (the primary player shot, drew a foul, or turned the ball over, or made a quick pass to a receiver who shot within one dribble of catching it)
  • New takeoff_feet and landing_feet fields on Jumps data set - Classifies which foot or feet the player took off from and landed on (left, right, or both; null when footwork cannot be determined)
  • New ballhandler_defender_initial_route field on Ball Screens data set - Classifies how the original ballhandler defender initially navigates the screen (over, under, or none)
  • New ballhandler_defender_switch_type and screener_defender_switch_type fields on Ball Screens data set - Classifies whether each original defender switched (normal/late) or stayed with assignment (none)
  • New screener_defender_initial_position field on Ball Screens data set - Classifies the screener defender's starting level in coverage (above, level, drop, deep)
  • New screener_defender_after_screen_defense field on Ball Screens data set - Classifies the screener defender's after-screen coverage (drop, drop_contest, switch, hedge, blitz, contain, bluff, none)
  • New screener_action field on Ball Screens data set - Classifies what the screener does after the ball screen (roll, short_roll, pop, mid_post, move_to_screen, none)
  • New rejected_screen and slip_screen_type fields on Ball Screens data set - Classifies whether the ballhandler rejected the screen and whether the screener slipped (none, quick, slow)

Fixed ​

  • Improved pass completion accuracy on Passes data set - fixed rare cases where passes were incorrectly marked incomplete when the intended receiver caught the ball and shot immediately before the opposing team gained possession.
  • Pass turnover classification on Passes data set - fixed rare cases where incomplete passes tied to a later dead-ball turnover were not marked as turnovers.
  • Incorrect splitting of possession touches on Possession Touches - fixed a rare issue where a possession touch could be incorrectly split into two separate touches.
  • Duplicate jump_shot_footwork_type rows on Jump Shots data set - fixed a rare issue where duplicate upstream shot-touch links could cause jump shot footwork outputs to fan out into duplicate rows for the same shot.
  • Bug in screener defender attribution on Ball Screens data set - fixed rare issue where the on ball defender was sometimes also marked as the screener defender.
  • False inbound tagging on Possession Touches and Passes data sets - fixed rare cases where live-ball out of bounds save plays could be mislabeled as inbounds passes and certain post-FGM inbound touches could show up as duplicate possession touches.
  • Improved classification of edge case in Advantage Creation 1v1 data set - improved handling of 1v1s where the floor is very static and the ballhandler doesn't really go anywhere but we can infer from the static floor that it was a 1v1 attempt.
  • Improved classification of corner movement and players heading out of bounds on After Shot Positioning data set - players in the corners who headed up the court at an angle are now separated better into actually getting back compared to positioning for an offensive rebound. Additionally, players headed out of bounds are more accurately classified, paritcularly those who just shot, and these players are not marked as positioning for an offensive rebound.

Changed ​

  • Expanded paint-touch qualification logic on Possession Touches, Chances, and Possessions data sets to handle edge cases.
  • Improved closeout_result classification on Closeouts data set - updated classification logic to improve accuracy in certain edge cases.

[0.3.0-alpha] - 2026-03-11 ​

Added ​

  • New ball_handler_loc_x, ball_handler_loc_y, and attacking_positive_x_basket fields on Advantage Creation 1v1 data set - Location coordinates of the ball handler at the start frame, along with a flag indicating which basket the offensive team is attacking
  • New screener_loc_x, screener_loc_y, ball_handler_loc_x, ball_handler_loc_y, and attacking_positive_x_basket fields on Ball Screens data set - Location coordinates of the screener and ball handler at the screening frame, along with a flag indicating which basket the offensive team is attacking
  • New screen detail fields on Ball Screens data set - screener_hoop_distance, screener_defender_hoop_distance_screen_start, screener_defender_hoop_distance_max, screener_defender_hoop_distance_diff, screener_defender_hoop_distance_max_diff, was_screener_defender_in_3pt_area_at_screen, did_ball_get_behind_screener_defender, screener_hoop_distance_min, ball_hoop_distance_min, ballhandler_outcome, screener_outcome, did_ballhandler_pass_to_screener, screen_distance_to_center_court, screen_distance_to_baseline, and did_ballhandler_dribble_after_screen for richer screen context and outcomes (distances in feet)
  • New Off Ball Screens data set
  • New chance_id_ctg and possession_id_ctg fields on Ball Screens and Off Ball Screens data sets - CTG-generated identifiers linking each screen event to its parent chance and possession
  • New paint-touch fields on Possession Touches, Chances, and Possessions data sets - Added has_paint_touch_during_possession to possession touches and total_paint_touches rollups to chances/possessions based on touches where the ball and both feet are in the paint.

Changed ​

  • Consecutive Chances no longer share the same start and end frame - When one chance ends and another begins immediately (rebounds, loose ball fouls, live ball turnovers), the new chance now starts one frame later to avoid frame overlap between consecutive chances.
  • Ball Screens end-frame timing now caps at ball-handler touch end - Ball screen windows now end no later than the end of the ball-handler possession touch that contains the screening frame, which prevents screen windows from lingering after the on-ball action has ended.

Fixed ​

  • Switch + roll handling in Ball Screens data set - In cases where a player set a ball screen, the defense switched, and then the screener rolled, they would sometimes end up facing the switched defender which would trigger our ball screen model to think they were setting a new screen. We have added more safeguards to better handle these cases.
  • Inbound possession touch robustness on Possession Touches data set - Improved handling of rare edge cases in inbound timing and touch boundary selection so outputs are more stable and consistent in ambiguous tracking/event scenarios.
  • Hoop-relative field accuracy across multiple data sets - Several hoop-relative measurements (rim distance, shot angle, contest distance, etc.) were occasionally computed against the wrong basket due to reliance on Hawk-Eye's attacking_positive_x field, which can be inaccurate at times. Basket side is now determined from a stable period- and team-level source, making these calculations consistent across Jump Shots, Jump Shot Contests, Free Throws, Shots, Interior Shot Contests, Closeouts, Contested Rebounds, and Jumps. Hoop-relative fields on Jumps will now return null when no reliable team context is available within 5 seconds of the jump, rather than returning a value computed against a potentially incorrect basket.
  • Period assignment on Rebounds data set - Fixed a bug where rebounds sometimes had incorrect period values in very rare cases at the end of periods; this had downstream effects on chance start and end frames.
  • Shot end frame calculation on Shots data set - Previously, when two shots occurred close together (e.g. blocked shots and putback shots), the end frame was incorrectly assigned to the end frame of the subsequent shot. We now cap shot end frames to the frame before the next shot starts in these cases.
  • Chance start timing (2D fallback) on Chances data set - During inbounds (particularly after long stoppages), ball and limb tracking can be highly error prone. Previously, we were using the first time we detected the inbounder touching the ball as the start of the chance in these cases, however in cases with bad upstream data this meant chances were starting later than they should have. This change makes it so we now use a 2D-based fallback to compensate in these situations, looking for when the ball is close to the inbounder's body. This may start chances too early sometimes, but this seems to be a more tolerable error than starting them too late.
  • Turnover frame timing on Chances data set - We found some cases where chance ends were incorrect on certain types of live ball turnovers. We have implemented a fix that should lower the error rate.
  • Inbound attempt deflected out of bounds included in Chances data set - When the ball is deflected out of bounds during an inbound (e.g. batted away before the pass is caught), we previously would miss categorizing these as chances. We now correctly treat this as a chance-ending stoppage. The chance end time aligns to the out-of-bounds event, and these sequences are handled consistently with other stoppages.
  • Clock flicker filtering on Chances data set - Added robustness to erroneous upstream clock stop/start flicker events by treating brief stop-then-start pairs (within 1 second) as noise and filtering them out.
  • Touch outcome for passes near lost-ball turnovers on Possession Touches data set - In some cases a pass we detected was actually part of a lost-ball turnover (e.g., ball batted away). We now prefer to assign the touch outcome as a turnover when a lost-ball turnover by the same player occurs within about 1 second of the pass, so the outcome matches the official result.

[0.2.1-alpha] - 2026-02-17 ​

Changed ​

  • Improved accuracy of height, wingspan, right_arm_length, and left_arm_length calculations on Anthro Measurements data set - Calibrated measurement constants using NBA Combine data to account for HawkEye skeleton changes in the 2025-26 season

[0.2.0-alpha] - 2026-01-13 ​

Added ​

  • New Ball Screens data set
  • New Advantage Creation 1v1 data set
  • New Possession Touches data set
  • New After Shot Positioning data set - this captures information on how offensive players are moving after each shot, including whether they were positioning for an offensive rebound
  • New chance_id_ctg field on Jumps data set - CTG-generated identifier linking each jump to the chance during which it occurred
  • New chance_id_ctg field on Drives data set - CTG-generated identifier linking each drive to the chance during which it occurred
  • New game_id_nba parameter on /game_status endpoint - Standardize NBA game ID naming to game_id_nba
  • New game_id_nba parameter on /markings endpoint - Standardize NBA game ID naming to game_id_nba
  • New game_id_nba field in /game_status response - Consistent NBA game ID naming in response body
  • New game_id_nba field in /markings response - Consistent NBA game ID naming in response body
  • New season parameter on /game_status endpoint - Filter games by NBA season (e.g., season=2025 for the 2025-26 season)

Deprecated ​

  • nba_game_id parameter on /game_status endpoint - Use game_id_nba instead
  • nba_game_id parameter on /markings endpoint - Use game_id_nba instead
  • nba_game_id field in /game_status response - Use game_id_nba instead
  • nba_game_id field in /markings response - Use game_id_nba instead

[0.1.0-alpha] - 2026-01-08 ​

Added ​

  • Initial creation of the CTG Pro Data Feed API documentation
  • New is_live_ball_rebound field on Rebounds data set - Indicates whether the rebound was secured as a live ball (not a team rebound or out of bounds). This is a clearer name for what was previously is_rebounded
  • New offense_players and defense_players fields on Chances data set - Arrays of NBA player IDs for the five players on offense and defense during each chance
  • New context and outcome fields on Chances data set - in_garbage_time (indicates if chance occurred during garbage time), shot_location (court zone where shot was taken), is_shot_made, is_shot_blocked, is_shooter_fouled, and led_to_offensive_rebound (shot outcome indicators)
  • New statistical fields on Chances data set - fga_pts (points scored on field goal attempt), fgm, fga, ftm, fta, tov, fg_dreb, and fg_oreb (four factors and basic statistics)
  • Standardized field names across markings data sets for consistency:

Changed ​

  • New heave option added to chance_outcome on Chances data set - Indicates an official heave

Deprecated ​

  • is_rebounded on Rebounds - Use is_live_ball_rebound instead. The old field name was confusing; the new name better describes the field's purpose
  • attacking_positive_x on Rebounds - This information is not particularly accurate and not useful in this data set
  • event_pbp_descriptor on Rebounds - This field is always null and provides no value
  • event_pbp_qualifiers on Rebounds - This information is redundant with other fields in the response
  • attacking_positive_x on Fouls - This information is not particularly accurate and not so useful in this data set
  • event_pbp_foul_personal_total on Fouls - This data is already available in the official NBA play-by-play feed
  • event_pbp_foul_technical_total on Fouls - This data is already available in the official NBA play-by-play feed
  • attacking_positive_x on Turnovers - This information is not particularly accurate and not useful for analyzing turnover events
  • event_pbp_descriptor on Turnovers - This information is not particularly useful for analyzing turnover events
  • event_pbp_qualifiers on Turnovers - This information is not useful for analyzing turnover events
  • event_pbp_turnover_total on Turnovers - This data is already available in the official NBA play-by-play feed
  • Deprecated field names replaced by standardized versions (all retain backward compatibility):
  • shot_clock_running on Turnovers - This field is unreliable and not particularly useful for analyzing turnover events
  • game_clock_running on Fouls - This field is unreliable and not particularly useful for analyzing foul events
  • shot_clock_running on Fouls - This field is unreliable and not particularly useful for analyzing foul events

Format and Versioning ​

The changelog format is based on Keep a Changelog.

It includes the following types of changes:

  • Added - New data sets, fields, or API features
  • Changed - Changes to existing data or behavior (may be breaking)
  • Deprecated - Features that will be removed in future versions
  • Removed - Features that have been removed
  • Fixed - Bug fixes in data accuracy or API behavior

The API follows Semantic Versioning (SemVer). Version numbers are formatted as MAJOR.MINOR.PATCH with optional pre-release labels:

ComponentWhen It ChangesExample
MAJORBreaking changes that require client updates1.0.0 → 2.0.0
MINORNew features added in a backward-compatible way1.0.0 → 1.1.0
PATCHBug fixes that don't change the API contract1.0.0 → 1.0.1

Pre-release Labels ​

  • alpha - Early development; API may change significantly between releases
  • beta - Feature complete; API is stabilizing but may still have minor changes
  • rc (release candidate) - Final testing before stable release

Versioning Policy ​

  • 0.x.x versions: The API is in active development. Breaking changes may occur between minor versions. We will notify you before this occurs.
  • 1.0.0 and beyond: The API is stable. Breaking changes will only occur in major version bumps, with advance notice in the changelog.