New York's Open Lottery Data Was Missing the $2.04 Billion Powerball Draw

On 7 November 2022 a Powerball ticket sold in Altadena, California won $2.04 billion — the largest lottery prize ever paid anywhere. When we checked New York State’s open-data portal in September 2026, it did not have that drawing. New York added it between our reads of its feed at 09:39 and 15:45 UTC on 30 September 2026.

Not “had it wrong”. Did not have it. The feed ran from 5 November straight to 9 November, with no gap marker, no null row and no note. Anyone building a frequency table from that dataset — and a lot of people do, because it is free, keyless and the deepest history on offer — was counting the history of Powerball with the most famous draw in the game’s existence quietly absent.

It was not the only one. Cross-checked against the Texas Lottery’s published exports and the game operators’ own results, New York’s feeds were short 9 draws and published 2 more with the wrong numbers: 11 defects in total, listed below with what has happened to each one since.

Where things stand

New York corrected 6 missing Powerball draws (7 November 2022 among them) between our reads of its feed at 09:39 and 15:45 UTC on 30 September 2026. Each one now matches the Texas Lottery feed exactly. New York corrected 3 missing Mega Millions draws between our reads of its feed on 30 September 2026 at 15:45 UTC and 1 October 2026 at 15:46 UTC. Each one now matches the Texas Lottery and Mega Millions (the operator) feeds exactly. We do not know what prompted the correction, and we are not claiming any credit for it. Still open, as of our last complete read of each feed: 2 Mega Millions draws with the wrong numbers (read 25 October 2026 at 04:37 UTC).

We first published this piece on 25 September 2026. The findings below are kept as we found them, each row marked with whether it is still true, because a correction is only checkable against a record of what was wrong.

Check it yourself, in about five minutes

Everything here comes from two public URLs that need no key, no account and no scraping.

Ask New York how many Powerball draws it has, and then ask it for that specific date:

curl "https://data.ny.gov/resource/d6yy-54nr.json?$select=count(*)"
curl "https://data.ny.gov/resource/d6yy-54nr.json?draw_date=2022-11-07T00:00:00.000"

The first now answers with a row count of 2,015. The second now returns the draw, 10 33 41 47 56, Powerball 10. Until 30 September 2026 it returned []: that empty list is what this piece was first written about, and seeing it filled is the quickest way to confirm the correction yourself.

Now ask Texas, which publishes the same game as a plain CSV:

curl -s https://www.texaslottery.com/export/sites/lottery/Games/Powerball/Winning_Numbers/powerball.csv \
  | awk -F, '$2==11 && $3==7 && $4==2022'

That returns the draw: 10 33 41 47 56, Powerball 10. At our last complete read of New York's Powerball feed, on 25 October 2026 at 04:37 UTC, it had 2,015 rows and nothing missing or wrong inside the window the other feeds cover. It was short 6 draws until New York corrected them on 30 September 2026.

The same two-line comparison works on Mega Millions (https://data.ny.gov/resource/5xaw-6ayf.json, and Mega_Millions/Winning_Numbers/megamillions.csv). At our last complete read of New York's Mega Millions feed, on 25 October 2026 at 04:37 UTC, it had 2,551 rows: wrong on 2 dates the other feeds agree about. Separately, it was short 3 draws until New York corrected them on 1 October 2026.

Every count on this page is as of our most recent complete read of each feed, and the time of that read is printed beside it. If New York changes a feed after that, what you get back will differ from what we print until our reconciler accepts the new version — and a read that suddenly returns far less history than the one before is held back rather than published.

Every defect we found

The 2 contradicted draws are worth looking at closely, because a missing row is an omission and a wrong row is an assertion. On both dates New York’s value was contradicted by two independent sources that agree with each other: the Texas Lottery’s export and the game operator’s own results service. In one case New York published a 31 where the other two have a 21; in the other, a Mega Ball of 06 against their 09. Those are single-character differences, which is what makes them survive — nothing about either row looks damaged. The Status column says whether New York still publishes them.

The arithmetic, in full

Our published totals are New York’s counts, plus the draws New York is missing, plus anything New York has not got to yet. Every term has its own column, and together they are why this piece can be falsified by arithmetic rather than by argument:

Mega Millions: 2,551 + 0 = 2,551. Powerball: 2,015 + 0 = 2,015, plus 1 draw New York has not published yet, for 2,016. Millionaire for Life: 244 + 0 = 244, plus 1 draw New York has not published yet, for 245. If our number is right, theirs is short by exactly the draws still open above. If our number is wrong, the same two URLs above will show it.

The last column is not an accusation and we have separated it deliberately. Feeds publish on their own schedules, and for a few hours after most drawings somebody is last; a draw newer than New York’s most recent row is a feed that has not caught up, not a gap. Only the column before it is a draw New York skipped and moved past.

What this breaks

Frequency analysis, mostly — which is the single most common thing built on this dataset. “Hot and cold numbers”, overdue-ball tables, gap charts, every “most drawn Powerball numbers” page: all of them are counts over a draw history, and a history with holes in it undercounts the balls that fell in the missing draws. Every table built on New York's Powerball feed before 30 September 2026 left out the balls that fell in the record drawing, 10 33 41 47 56, Powerball 10 — and one computed then and not rebuilt since still does. The feed's correction reaches nobody who already downloaded it.

We think that particular effect is small and we want to say so plainly: a handful of draws out of two thousand moves a frequency count by a fraction of a percentage point, and anybody claiming a ball is “due” was wrong before the gap and is wrong after it. The problem is not that the numbers move much. It is that a table built from one feed cannot tell that anything is missing, and neither can its readers. The feed does not report its own gaps, so a site built on it inherits them silently and publishes them with the same confidence as everything else.

What we are not claiming

Five things, because a piece like this is worth less if it overreaches.

We are not claiming New York’s data is bad. Its Mega Millions archive reaches back to May 17, 2002, deeper than the operator’s own feed and deeper than Texas’s. It is free, well-documented, fast and keyless, and it is the best starting point in the country. It has also had holes in it, and both things are true. Here is every feed behind the two games, and the window each one covered at our last complete read:

We are not claiming we cross-check everything. We do not. Of the 139 draw games on this site, 11 are published by more than one official body — Pick 3, Pick 4, Fantasy 5, Pick 2, Pick 5, Florida Lotto, Mega Millions, Powerball, Jackpot Triple Play, Millionaire for Life and Lotto America — and those are the only ones whose draws we can compare against a second source. The other 128 have exactly one official publisher, so every draw in them is that publisher’s record with no second opinion available, and we would rather say so than imply a check we never performed. Our sources page names the feed behind every single game, and each game’s archive page says it again.

We are not claiming Texas is the truth. Texas is a second reading, which is the only thing that made any of this visible. Where the two disagree about numbers rather than about whether a draw happened, we defer to the game operator’s own results — closest to the drawing itself — rather than picking a favourite state, and we say which we did on every row.

We are not claiming the state hid this. The dataset’s own metadata is upfront that it is not the authoritative record: it names the Gaming Commission’s Drawing Unit files as the ones that prevail in any discrepancy, and it gives a contact address for exactly this kind of report — Info at gaming.ny.gov, written out that way because our own host rewrites anything that looks like an email into a placeholder, and a page about being checkable should not print one. Nobody reads a dataset’s notes field, which is rather the point — the feed is treated as authoritative because it is official, free and easy, and the disclaimer does not travel with the JSON. So this is not a story about a cover-up. It is a story about what happens to a caveat when a dataset is convenient.

We are not claiming to know why. We have no theory about the cause. The gaps did not share an obvious shape: they spanned 2007 to 2022 and fell on ordinary and extraordinary drawings alike. The record jackpot draw was delayed several hours by a participating state’s security check and was conducted the following morning, which is the kind of thing that breaks an import — but we have not verified that is what happened here, and we are not going to imply it.

What we did about it

We publish the union. Every draw either feed carries is in our tables, every disagreement is adjudicated in favour of the operator, and every one of those decisions is listed draw by draw on our sources page rather than described in the abstract. A game whose feeds disagree in a way no operator can settle is not published at all until it is resolved, and that rule is enforced by the build rather than by anybody remembering to check.

None of the numbers in this piece were typed. They are interpolated from the same export that builds the Powerball archive, the Mega Millions archive and the sources page. When New York repairs a draw, our next read moves it from open to corrected on this page by itself, dated by the reads either side of the fix, rather than leaving a stale accusation standing or quietly deleting the record of it — and if a new gap appears, the page grows without anyone remembering to edit it. How we compute everything is written up separately.

If you find an error here, tell us — the official page is right and we want to know.