After the NBA Moved Its Injury Report to Every 15 Minutes, How Much News Lands Off the Hour?
We counted each player’s first status change on the official report during the 2025-26 regular season (from December 22, 2025) by the report it first appeared in (:00, :15, :30 or :45), to see how much news you would see late if you only checked on the hour.
- Result status
- Preliminary results
- Last updated
- Start date
- Duration
- Day 4 of 4
- Research team
- Mysports.AI Research Team
- Plan registered
NBA Injury Report Timing: Key Findings
3,280 changes in the regular season from December 22, 2025 to April 12, 2026 (71.9%, range 69.2%–74.6%). An even split would be about 74.9%, so the hour gets only slightly more, and that is not stable across months.
- By report: :00 28.1%, :15 26.0%, :30 21.4%, :45 24.5%. The hour is about 3 points above an even split, :30 about 4 below, and no difference can be seen for :15 and :45.
- Checking on the hour: about 72% seen late, by a median of 30 minutes and at most 45 for those. Checking every 30 minutes (:00, :30) still sees about half late, by a median of 15 minutes.
- Very little stays hidden until after the scheduled start: 0.2% when checking on the hour. But 13.1% only show up in the report at the start time or later, since many games start on the hour.
- News comes late: 54.8% within 2 hours of tip-off, 44.7% in the 45–90 minutes before it (a sum added after seeing the results), and only 72 changes the day before or earlier.
- Teams’ first submissions are more concentrated on the hour (41.6%), which fits the official deadlines all being on local whole hours, though fitting is not proof.
- No difference can be seen between key players (71.7%) and others (72.1%), and changes to out, to available and to in-between are all around 70%. Play-in and playoff games have only 142 changes, 80.3% off the hour, for reference only.
In plain words
Why study this?
The NBA’s official injury report now comes out 96 times a day, and our own data collection fetches it every 15 minutes. So we asked: is the news really spread across every report, or does most of it still land on the hour, so that checking on the hour would be enough?
What do we want to know?
Before every game the NBA publishes whether each player can play. From December 22, 2025 the official report went from hourly to every 15 minutes, 96 times a day. We ask: is the new information really spread across all the reports, or does most of it still land on the hour, so checking on the hour would be enough?
How do we study it?
Using the official report PDFs, we looked at each player’s first status change for each game from December 22, 2025 to April 12, 2026 (for example from “questionable” to “out”) and noted the first report it appeared in. That gives 3,280 changes, all before tip-off and exactly 15 minutes after the previous report. We then simulated different checking schedules to see how much would be seen late. This study does not look at odds. The plan and the analysis code were timestamped before any results were computed.
What have we found?
71.9% of first changes land in the :15, :30 and :45 reports and only 28.1% on the hour. An even split would put about 25% on the hour, so the hour gets only slightly more, and that is not stable across months (22.9% in March, 34.1% in April). The :30 report is actually the lowest (21.4%). If you only check on the hour, about 72% of the news is seen late; for those, the median delay is 30 minutes and the longest is 45. More than half (54.8%) of first changes come within 2 hours of tip-off, with 44.7% in the 45–90 minutes before it (a sum added after seeing the results). We see no difference between key players and others.
How does this help with predictions?
To learn news as soon as possible you should check every 15 minutes: checking hourly sees more than 70% of the news late, and checking every 30 minutes still sees about half late. To check less often, you could consider every 15 minutes within 3 hours of a game and hourly otherwise, which sees 21.7% late. A model that uses the report only at a fixed time (for example 2 hours before tip-off) does not, in terms of information, need constant fetching, only the one report it uses at that time, though we have not tested that full workflow. More importantly, more than half of first changes come after that 2-hour mark, so no fetching frequency can help a prediction made 2 hours before the game. This is not betting advice.
What can’t we say yet?
This is a retrospective study: the reports are public and our earlier study R1 used the same reports, so it is not a blind test, though the share by report was never computed before the plan was timestamped. The time is when the official report was posted, not when the news actually spread; reporters may have said it earlier. We cannot separate team submission habits from how the league compiles and posts reports. This is one season and under four months. Games mostly start on the hour or half hour and news clusters at a certain time before tip-off, so the shares are affected by start times, and they alone cannot show that teams rush for the hour. For 2026-27, we will recalculate with the same rules after R1’s forward cutoff.
Terms used on this page
- Injury report
- The NBA’s official list saying whether each player can play in a game: Out, Doubtful, Questionable, Probable or Available.
- First status change
- The first time a player’s status for a game differs from the previous report. A team adding a player it had not listed also counts (not listed is treated as available). A team’s first submission does not count, and neither do later changes.
- Off the hour
- The reports at :15, :30 and :45 past the hour; on the hour means the :00 report. Times are US Eastern, but all US and Canadian team time zones differ by whole hours, so the minutes are the same.
- Even split
- The share each report would get if every report and every team were equally likely to carry news, about 25% each. It is only a yardstick, not a claim about how news should be spread.
- Range, percentage point
- The range is a 95% interval from resampling dates 10,000 times. A percentage point is the difference between two shares, for example 28% and 25% differ by 3 points.
Detailed results
How much news is seen late with each checking schedule
3,280 first status changes in the regular season, December 22, 2025 to April 12, 2026. Numbers are the share seen late.
The pre-registered main result is the first row (range 69.2%–74.6%). Our pre-set rule calls a lower frequency safe only if the upper limit of the share seen late is below 10% and of the share not visible before the start is below 1%. None qualifies, so to learn news quickly, keep checking every 15 minutes. This is research, not betting advice.
HypothesesFor deeper reading
- Q1More than half of first changes land off the hour → supported: 71.9% (range 69.2%–74.6%) appear in the :15, :30 and :45 reports, and this holds under every alternative we triedSupported in this sample
- Q2The on-the-hour report gets more than an even split would give → supported under our pre-set rule, but only by about 3 percentage points (28.1% vs about 25.1%), and this is not stable across methods or months. It does not mean the news is concentrated on the hourSupported in this sample
- Q3The :30 report gets more news (a habit left over from the old :30 snapshot) → not seen: :30 is actually the lowest (21.4%, 3.8 percentage points below an even split, range 1.5–5.9 points). This does not prove the old habit is entirely goneNot supported in this sample
NBA Injury Report Timing: Research Methods
Research questions
- What share of first status changes appears in the :15, :30 and :45 reports rather than the on-the-hour one?
- The old hourly report was really a snapshot at :30 past each hour. After the change, does the :30 report get more news?
- If you only check on the hour, how much news do you see late, how late, and is any of it not visible until after the game starts?
Method summary
- 15-minute files actually start on December 22, 2025 at 09:00 ET. The main sample runs to the last regular-season day, April 12, 2026: 3,280 changes on 105 dates. Apart from the five-day All-Star break (the official server would not open those days’ files, and no games were played), every slot has a file, and the four slots have exactly the same number of files.
- There is one main result: the share in the :15, :30 and :45 reports. The range resamples whole dates, because changes in the same team report always share a slot and teams on the same day are not independent. Checks with other groupings did not change the wording against 50%; the comparison with an even split is not robust under some of them.
- The old hourly report was really a :30 snapshot (file named 05PM, content stamped 05:30 PM), so the comparison with the old system looks at :30, not the hour.
- The first version of our pre-set data gate did not pass: the official server would not open files for the five-day All-Star break, and one day’s files (March 26, 2026) failed a check because a next-day game’s start time was blank. The lab reviewed this and revised the gate before any share by slot had been computed. Removing March 26 entirely gives 71.8%.
- All data checks passed, including 30 of 30 spot-checks against the PDF text. The plan had two rounds and the results one round of external code review; the analysis code was timestamped before running and not changed afterward.
Data and limitations
Data we use (all public and free)
- NBA official injury report PDFs (2025-26 season; this page shows only aggregated numbers)
- NBA game-by-game minutes (nba_api; only to separate key players from others)
Limitations
- Retrospective, not blind: the reports are public and we had used them in an earlier study.
- Times are when the report was posted, not when the news actually spread.
- Team habits and league posting cannot be separated, and game start times also affect the shares.
- One season and under four months; the on-the-hour share varies a lot by month.
- Treating an unlisted player as available is an assumption; without those changes the result is still 71.9%.
- Simulations use scheduled start times; actual tip-off is usually a few minutes later. This is not betting advice.
Latest progress
Latest：Main results are in: 71.9% of first status changes land off the hour, and the hour gets only slightly more than an even split; to learn news quickly, keep checking every 15 minutes. The plan and analysis code were timestamped first, and results were externally reviewed
Next step：For 2026-27, recalculate with the same rules after R1’s forward cutoff
Study timeline
- First version of the data gate did not pass; revised and passed after review, before any share by slot was computed
- Plan finalized and timestamped after two external code reviews; analysis code timestamped before running
- Main results in; one round of external code review (main number unchanged)
- For 2026-27, recalculate with the same rules after R1’s forward cutoff (1,000 items or February 15, 2027)In progress
Update log
Version history
- v1.0-results
Main results: 71.9% of first status changes land off the hour; checking on the hour sees about 72% late; to learn news quickly, keep checking every 15 minutes
- v1.0
Plan finalized and timestamped (T0 = 2026-11-05 03:57 UTC); retrospective, not blind, with 2026-27 as a later re-check
Study record
ac99f267eed0d79d024a4b78506429a92cd6804f3e416daf163fd5348b8ed15a741ec604afd4bd2691680d514b2cb173f570b79a5cca8eccfeda929faddcede1f78db968e0f881bd639c1ced429424b9e5c21f28c7d66174b9944f9bcea66cb92ef87f701265dc195e81933eec466c6ee76218f3f964e4864334ebbf30edf170This page is a research log for reference only and is not betting advice or a guarantee of profit.