Data and limits
Data sources, season coverage, exclusions, and known gaps in FullCourt's analyses.
Keep in mind Coverage differs by field and season. An unavailable input is not evidence that no overtime, travel or other event occurred.
On this page
Data sources and coverage
42 seasons; coverage varies by fieldCoverage depends on the field. The table lists each source and the earliest coverage used here; individual studies apply further sample filters.
| FIELD | FROM | SOURCE | NOTE |
|---|---|---|---|
| Player box scores | 1996-97 | NBA / hoopR | Player-level shooting and availability inputs; each analysis applies its own filters. |
| Last Two Minute reports | 2014-15 | NBA Official | Published regular-season reports through 2025-26; selected close-game endings only. |
| Schedules, scores, results | 1985-86 | NBA feeds | Complete across every covered season. |
| Overtime periods | ~2002 | ESPN | Earlier seasons read 0 because overtime is unknown. |
| Tip-off times | ~2002 | ESPN | Without one, the back-to-back multiplier falls back to its flat value. |
| Neutral-site venues | 2013 | ESPN | Earlier international games are scored at the listed host's arena. |
| Shot locations | 1996-97 | hoopR / play-by-play | Expected Shot Value only. These records do not include defender or tracking fields. |
Input coverage changes around 2002. The same formula uses the fields available for each game, so a 1994 score and a 2024 score are not built from identical information.
Exclusion rules and their history
| Games | Rule |
|---|---|
| Playoffs | Excluded from regular-season fatigue analysis; modeled separately in Playoff Rest. |
| Orlando bubble | Excluded; the pre-suspension part of 2019-20 remains in regular-season analysis. |
| 2019-20 in Schedule Edge | Entire season excluded from rankings because teams played unequal schedules. |
| Preseason | Excluded; rotations do not represent regular-season play. |
Playoff games are excluded from the regular-season fatigue model because a fixed two-team series breaks its travel assumptions: the opponent never changes and the itinerary is known weeks ahead. They are modeled as whole series on the Playoff Rest page.
The 2019-20 Orlando bubble is excluded because every game was played at a single site: there is no travel to measure and no home crowd to weigh. That is a statement about roughly 88 games between 30 July and 11 October 2020, not about the season. The 971 games that season played before the 11 March suspension were reached by flying to them, and they are in.
Exclusion correction: July 2026
Until 30 July 2026 the whole of 2019-20 was absent, and those ~970 ordinary games went with it. The rule was written as an October-to-April calendar window, which caught the bubble only by coincidence of dates, and along the way dropped 179 legitimate games from seasons that did not run October to April: 135 from 2020-21, which ran to 16 May, and 44 from the 1998-99 lockout season. The exclusion now uses the bubble dates directly.
One surface still withholds the season in full. Schedule Edge ranks teams against each other within a single season, and 2019-20 stopped with teams having played between 63 and 67 games. A team with four fewer games has four fewer chances to accumulate an edge, so its total would move without the schedule having favored anyone. This exclusion addresses unequal schedule lengths and applies to no other season on record, where the widest spread is a single game.
The lockout seasons are kept for the same reason 2020-21 is: 1998-99 ran 50 games and 2011-12 ran 66, both with normal travel and both complete. A short season is only a problem for this site when it is also interrupted.
How travel is estimated
The game log records cities, dates, and game order. The model assumes the itinerary between them: teams are modelled as flying venue to venue, returning home only when the next game is at home. No public itinerary data is used to verify those assumptions.
Distances use great-circle routes. Actual flight paths, stops at home, and recovery time can differ from this estimate. The data here cannot quantify how those differences affect a team’s fatigue.
What each model includes
The regular-season fatigue score uses schedule inputs and excludes team strength, betting lines, and expected lineups. Playoff Rest uses a separate series model that includes regular-season team records. Availability cost uses team strength as a control and identifies missing players from completed games.
Historical rest splits describe associations. They do not isolate the causal effect of rest or account for every factor needed to forecast a game.
Known data gaps
- —Pre-2002 overtime is unknown rather than zero. Basketball-Reference carries it and could close this gap; it has not been done.
- —Neutral-site games before 2013 are unmarked, so Mexico City games from 1997 and the 2011-12 London games are geolocated at the listed host's arena. Basketball-Reference marks them.
- —Arena coordinates are era-correct for relocations, but neutral-site venues rely on a small hand-maintained list of cities.