Every other case on this shelf is about somebody else's mistake. This one is ours.
South Africa's PowerBall is five numbers and a powerball. For sixteen years, this database held it as five numbers and nothing else.
Not some of the draws. All of them, from 2010 until June 2026 — 1,707 results in which the sixth ball, the one the game is named after, was simply absent. The draw of 1 December 2020 that closes the previous case was stored here as 5, 6, 7, 8, 9 and a blank.
The Missing Ball
The draw of 1 December 2020, as this database held it for years, and as it happened.
Five numbers, and nothing where the powerball goes. Every check passed: the numbers were in range, sorted, the right count for a game the feed had described as having no bonus ball.
Nothing here reported an error. The feed returned success, the draw validated, and the game simply looked like one that had never had a powerball — which is what makes a silent field harder to catch than a broken one.
What makes this worth a case file is not the hole. It is that nothing anywhere in the pipeline said a word.
The commercial feed that supplies most of this site returned those draws with five numbers and a null where the powerball goes, and reported success, because from its side nothing had failed. The records validated, because a rule era had been derived from the records themselves: the deriver read sixteen years of five-number draws, concluded the game had no bonus ball, and wrote that down as the rule. From then on the data was not merely wrong, it was consistent. Every check compared the draws to a description of the game that the draws had written. They agreed.
There was even a decoy. The same feed carried PowerBall Plus — the companion draw on the same night, from the same machine — and Plus had its powerball throughout. Any glance at the game showed powerballs present and correct. They were just in the wrong half of it.
The operator's own site answers automated requests with a 403, so the fix came from the year archives of a results site: 1,749 draws from October 2009 to August 2026, every one carrying exactly one powerball between 1 and 20. Before trusting a word of it we checked it against what we already had. On the 379 draws the two sources share, the five main numbers agree 379 times out of 379, and on the 22 draws where the commercial feed did carry a powerball, it agrees there too.
And then the second source did nothing, twice, because of two more faults of our own.
The first was in how agreement is decided. Two sources "agree" when they report the same draw, and the code compared the main numbers and the bonus together — so a five-number reading and a five-numbers-plus-powerball reading of the same night were not the same answer told with different detail. They were rival answers, and the terser one had more votes. The fix is a sentence long and took a week to see: a source that omits a field is not disagreeing about it. It is abstaining, and the informative reading should lead.
The second was in the era deriver. Asked for one record per date, it took the one with the lowest source id — which was the older, terser feed. So even with the archive loaded it would have gone on reading five-number draws, gone on concluding there was no bonus ball, and then refused every record that had one, as malformed. A game can acquire a second source and still show nothing.
With both fixed: 1,729 of 1,729 draws promote, 1,707 of them gaining a powerball they had never had here, plus twenty draws from 2009 that predate the commercial feed entirely. The rule eras that fall out of the corrected data are the game's real ones — five from forty-five with a powerball from twenty, and five from fifty from June 2018.
We publish this because a site whose entire argument is that it reports what happened owes the reader its own errata, and because the shape of this one is worth recognising. A feed that breaks loudly is fixed the same morning. A feed that goes quiet in a single field can be believed for sixteen years, especially once the record has been allowed to define the rule it is checked against.
The scan we run every night now counts the draws it refuses. It cannot yet count the ones nobody sent.