Every pick we publish is written to the ledger before the draw and graded after it. Alongside six selection strategies we run a seventh set chosen with no logic at all — a control. Across 41,727 settled picks we cannot detect any difference between our strategies and that control. This page shows the data, the method, and the SQL to check us.
Random Baseline uses no frequency analysis, no hot or cold numbers, no model. It is the scientific control, and it exists to answer one question: does the analysis beat picking at random?
Running a control is uncomfortable, because it can only ever embarrass you. We publish it anyway. Below is every strategy and tier, ranked by return, with the control in gold.
The control lands mid-pack, beating several strategy-and-tier combinations outright. If the models carried information, this is exactly where it would appear. It does not.
Two cells made money. Before drawing any conclusion, look at what a single ticket does to each.
A Pick 4 straight pays $5,000 on a $1 ticket. One hit is worth more than twice a strategy's entire annual spend. In 41,727 settled picks, nothing else won more than $500. These are not patterns. They are two lottery tickets, and hitting one is a random event.
With twenty-one strategy-and-tier combinations buying tickets daily, a couple of big hits landing somewhere is what probability predicts. Naming the cells they happened to land in and calling them winners is how a random outcome gets sold as a system. We won't do that.
Settled picks only, graded against the official draw at each tier's native bet. Identical sample and identical spend in every cell, so the columns compare directly.
| Strategy | Tier | Wins | Win rate | Spent | Won | Return | Best hit |
|---|
Win rates span —. The control sits inside that band, not below it. Aggregate return across all 41,727 settled picks: —.
Which numbers you pick cannot change your expected value. How you bet them can — slightly. Florida publishes its payout tables, so this is arithmetic, not opinion.
| Structure | Pays | Odds | EV per $1 | True ROI |
|---|---|---|---|---|
| Pick 4 · Straight | $5,000 | 1 in 10,000 | $0.50 | −50.0% |
| Pick 4 · Box (24-way) | $200 | 24 in 10,000 | $0.48 | −52.0% |
| Pick 3 · Straight | $500 | 1 in 1,000 | $0.50 | −50.0% |
| Pick 3 · Box (6-way) | $80 | 6 in 1,000 | $0.48 | −52.0% |
The true gap between straight and box is about two percentage points. Replay our own ledger and straight shows −38.8% against box at −66.4% — a gap of twenty-eight. That spread is not an edge. Straight is heavy-tailed: we expected 0.23 jackpot hits and got two, dragging the observed figure far above its true −50%. Box hits often and small, and still lands well below its true −52%.
Neither number has converged. Read the replay and conclude "always play straight" and you are acting on noise — the same trap the strategies fall into, one level up. We show the replay because it is genuinely interesting. We show this table so it cannot be misread.
Selection strategies cannot win because the thing they model has no memory. These five tests run in your browser, right now, against our live draw archive. Reload and they re-run on fresh data.
A note on method. These tests run within a single game's own chronological history — never pooled across games. That distinction is the whole ball game. Pick 2 sums span 0–18; Pick 5 sums span 0–45. Because draws arrive ordered by date, a pooled series lurches between different distributions and manufactures textbook "correlation" and "clustering" out of perfectly random numbers. Nearly every lottery "pattern" claim you will read online is built on exactly this error. Feed a time-series test a series that isn't one, and it will hand you a pattern every time.
A Pick 4 straight pays $5,000 on odds of 1 in 10,000. All ten thousand combinations carry the same expected value: fifty cents on the dollar. No subset has better math, because the payout is fixed and the draw has no memory. Hurst exponents, cointegration, mean reversion, RSI — every one of those tools exploits dependence between observations. Here, there is none.
Draw games cannot be beaten by selection. Three things on this platform are arithmetic rather than marketing, and they are the reason it exists.
A scratch-off is a finite pool, not a fresh draw. Once the top prizes are claimed the remaining tickets are objectively worth less — and Florida publishes how many are left. That is a real, observable, decaying signal. Our engine tracks it per game and per price point. It will not make a scratch-off profitable; it will tell you which ones are worse than they look, and which have been quietly gutted.
Anti Split does not raise your odds of winning — nothing can. It raises your expected share of a jackpot by avoiding birthdays, dates, and the patterns thousands of other players use. Identical probability of hitting; fewer people to divide with. A real edge on payout rather than odds, and the only strategy here with a defensible mechanism behind it.
Powerball's expected value genuinely moves with the rollover. There are jackpot levels where nominal EV crosses positive — before tax and split risk claw it back. Knowing where that line sits, and how rarely it is truly crossed, is worth more than any number-picking system.
Every figure in section 03 comes from one query against the forward ledger. Nothing is curated, back-fitted, or filtered. Picks are written before the draw; results are graded after it.
select p.strategy, p.risk_tier, count(*) as settled, sum(case when r.prize > 0 then 1 else 0 end) as wins, sum(p.ticket_cost) as spent, sum(r.prize) as won, round(((sum(r.prize) - sum(p.ticket_cost)) / sum(p.ticket_cost) * 100)::numeric, 2) as roi_pct, max(r.prize) as biggest_win from forward_picks p join forward_results r on r.pick_id = p.id group by p.strategy, p.risk_tier order by roi_pct desc;
Some ledger rows are tagged backfilled. In July 2026 a maintenance job deleted ledger history older than thirty days. We regenerated the missing period from the same inputs, using only draws that occurred before each pick date, so no future information entered any pick. Those rows carry the tag, and we say so here rather than folding them in quietly.