A place for the weird stuff
Balliqa's main product is useful: free data-driven picks, a number checker, and a scoring audit that shows exactly how the model behaves. Those pages have a job. They're not the place to ask "what does a Powerball draw sound like if we turn each number into a note?"
So we made a place for that too. The Lab is a hub for experiments — little self-contained pages that look at the same 1,929 historical draws from unusual angles. Some are pure visualization. Some are randomness tests. One is an A/B experiment that produces a real result. None of them predict anything.
Here's a tour.
The eleven experiments
Combinatorial vs Empirical — our live A/B test. Two scoring models (the current combinatorial v6.0 and a resurrected empirical v3.1) score the same 50,000 random candidates every draw day. The ten top picks from each go into the database before the draw. A month in, neither model has a significant edge — which is the outcome combinatorial theory predicts.
Collatz × Powerball — every number from 1 to 69 traces a unique path to 1 under the Collatz rule (halve if even, triple-plus-one if odd). We render all 69 paths with the latest draw highlighted. Purely deterministic, purely decorative, and weirdly beautiful.
Draws on a Clock — place 1–69 evenly around a circle, connect the five drawn numbers into a polygon. The centroid color signals how symmetric the pick is: green for near-pentagon, red for lopsided. About 1 in 8 draws looks roughly symmetric.
Draws as Colors — a 200-tile wall where each tile is one drawing. Hue comes from the circular mean of the five numbers, saturation from spread, lightness from sum. A streak of dark-blue tiles means the lottery quietly favored low numbers for a week. (It didn't. But the wall lets you verify.)
Draws in Binary — every draw is 35 bits (5 numbers × 7 bits each). We render the last 60 as bitmap rows, then plot each of the 35 positions' 1-rate against the uniform-1-to-69 expectation. The bit-flip rate between consecutive draws is 50.0%, exactly as independence predicts.
Modular Fingerprints — residues mod 2, 3, 5, and 7 for every ball in every draw. By CRT these four residues uniquely identify each number 1–69 (since 2·3·5·7 = 210 > 69). The composition plots show observed residue frequencies landing within sampling error of expected. Boring! That's the point.
Prime Factorizations — every drawn number broken into prime factors, plus a frequency bar for each prime across all draws. 2 dominates because half of 1–69 is even. 67 barely appears because it's prime and its only multiple in range is itself.
The Space Between — forget the numbers. Every draw is really six gaps that add up to 69 (before the first ball, between each pair, after the last). The histogram of internal gaps skews toward small values — and about 27% of all draws contain at least one adjacent pair, a number that surprises most people.
Draws as Chords — map 1–69 to 69 chromatic semitones starting at C3. Tap a recent drawing, the browser synthesizes five sine waves at those frequencies into a chord. Most draws sound dissonant because random semitones rarely land on a scale. Headphones recommended.
Is Your Draw in π? — type five numbers, we search the first million digits of π for the concatenated 10-digit string. Below the input, the twelve real draws whose prefix appears earliest in π. More on this one in a moment.
RNG Tournament — eight random number generators (Mulberry32, Xorshift128, LCG, SplitMix, MSWS, PCG, Fibonacci, Math.random) generate 50 picks per historical draw. Each's best match scores. Over 1,929 draws, no generator pulls meaningfully ahead. The actual tournament-winner column shows each RNG winning roughly 1/8th of the time — pure noise.
The technical one: how π tells us nothing useful
The π search is the experiment that feels like it should mean something but doesn't. Let me walk through why it's interesting anyway.
Concatenate a draw's five balls into a 10-digit string (7 becomes 07, etc.). The chance that any specific 10-digit string appears at any specific position in a uniformly-random digit sequence is 10⁻¹⁰. The chance it appears somewhere in the first N positions is:
P = 1 − (1 − 10⁻¹⁰)^N
For N = 10⁶ (the first million digits we search), that's ≈ 0.01% — essentially zero. So when you type a draw and the full 10-digit string isn't found, that's the correct, expected answer.
But π is suspected to be a normal number: every finite digit sequence appears with the frequency you'd expect from a uniform source. Nobody has proven π is normal, but every statistical test ever run on it — up to the first 100 trillion digits computed in 2022 — agrees. So your numbers are somewhere in π. Just somewhere beyond position 10¹⁰.
The experiment's fallback is more satisfying. When the full string isn't in our million digits, we search for the longest prefix that is — 8 digits, 6, 4, 2. A 4-digit match (first two balls concatenated) is almost guaranteed by the same math: with 10⁴ possible strings and 10⁶ positions, expectation is 100 hits. So the "earliest appearing" table always has entries.
The practical lesson: π is a wonderful source of entropy, but it's also a reminder that "your numbers exist somewhere in this famous sequence" and "your numbers were more likely to appear in the last drawing" are both true statements that tell you exactly nothing about your odds. Every 10-digit combination is equally likely to win. Every 10-digit combination exists somewhere in π.
Why ship this
Two reasons.
One: visualization is a form of verification. We spend a lot of words telling readers that Powerball draws are independent random events, that past frequency doesn't predict future draws, that no system can shift the odds. Putting twenty different lenses on the same data and showing that none of them reveal a hidden signal is more convincing than any paragraph. The randomness page makes this case formally. The Lab makes it visually.
Two: it's fun. Random numbers are among the most fertile material in math — Collatz sequences, modular arithmetic, spectral tests, normality conjectures, RNG theory, Monte Carlo simulation — and a lottery dataset is an excuse to touch all of it without pretending any of it will win you money.