Basement Whatever

Can AI be used to analyze Powerball data? With the right dataset, it sure can.

First I grabbed a list of winning Powerball numbers from data.gov. The name of the dataset is “Lottery Powerball Winning Numbers – Beginning 2010”

Then I tried a technique I learned at work. Rather than flailing around trying out prompts, after I loaded the data into Claude, I asked for a prompt:

What are good prompts I can use to query this data?

I tried “What are the 10 most frequently drawn winning numbers”. The answer is 28, 23, 39, 36 and 32, in that order. Claude says the numbers are clustered tightly, indicating an even distribution, but I asked:

Is there anything about the distribution that is unusual?

SURPRISE! Claude said “there are two genuinely interesting anomalies in this data”. 1/ The bottom 10 least frequent numbers are almost entirely in the 60-69 range. This is a real statistical outlier. Then I learned something else: According to Claude, the reason is structural: Powerball expanded is main ball pool from 1-59 to 1-69 in October 2015, meaning numbers 60-69 haven’t had as many draws to appear in.

NEXT SURPRISE! Claude sees discrepancy here:

Numbers 1–26 appear roughly 60–80 times each, while numbers 27–39 appear only 3–27 times, and nothing above 39 exists at all. This is because in October 2015, the Powerball pool was shrunk from 1–39 down to 1–26. Numbers 27–39 only had a few years of draws, while 1–26 have the full dataset….

In short, the “hot” and “cold” numbers aren’t really about luck — they’re almost entirely explained by the October 2015 rule change that restructured the entire game. 

Fair enough, but then:

Given number 2, any explanation for why 28, 39, 36 and 32 are in the most frequently occurring numbers?

Claude explains that this is attributable to the rule change. Even though 2015 expanded the number of balls to 69 from 59, each individual number gets picked less often per draw when the pool expands.

Now I figured, what happens if we try some weird numerology stuff? I asked:

When adding up the winning draws, can any conclusion be drawn from the sums?

Oops, I’m not as clever as I thought. That is actually just a bell-curve answer, and the results follow what probability predicts. Most results are in the 150-174 range, followed by 175-199 and finally 125-149 which is statistically normal.

Now I’m really desperate, wanting something like the movie Pi to jump out. So I go for a leap:

Is there anything weird about these values? Anything that maps to common numbers in astrology for instance?

Claude states that “This is a fun rabbit hole” but “The honest answer: no, not really”. A few tidbits are that the Angel Number 333 and Tesla’s famous “sacred number” 369 have never been a winning sum, but that’s because it would require five very high numbers which is “rare territory.” Also, only a single draw in 15 years had 4 of 5 numbers as Fibonacci numbers (on May 26, 2021). According to Claude:

The bottom line is that the subs heave like a textbook normal distribution…if astrology or numerology were influencing the outcomes, we’d expect certain sums to cluster abnormally, and they simply don’t.

In conclusion, nothing exciting like the clip below, but I tried!