Hi-lo guide
Ruby Hi-lo: reading a seeded simulation
The word Ruby in this address is a leftover label; it does not name a different deck or payout. This page runs a small computer simulation of hi-lo and explains how to read it without over-reading it.
What is simulated
Each simulated call shows a Jack and guesses lower. Ten ranks (Ace to 10) beat that guess, so the win chance is 76.92%. The next rank is drawn uniformly from 1 to 13 by Python's seeded pseudo-random generator. A win pays 1.2870x of a one-unit stake (a 1% edge); a loss costs the unit. There are six sessions of 120 calls each.
| Session | Rounds won | Net (units) | Lowest point | Longest losing run |
|---|---|---|---|---|
| 1 | 92 | -1.60 | -3.18 | 4 |
| 2 | 94 | 0.98 | -4.35 | 3 |
| 3 | 96 | 3.55 | -3.14 | 2 |
| 4 | 102 | 11.27 | -0.43 | 2 |
| 5 | 86 | -9.32 | -9.32 | 2 |
| 6 | 95 | 2.26 | -2.14 | 3 |
Seeded pseudo-random simulation (Python random.Random, seed 7741), 6 sessions of 120 one-unit rounds. Average session result 1.19 units. This is a simulation, not a record of real play; a different seed gives different numbers.
How to read each column
- Rounds won
- How many of the 120 calls won. The exact expectation is 120 x 0.7692 = 92.3 wins.
- Net (units)
- The total result of the session. The exact expected net is 120 x (-0.01) = -1.20 units.
- Lowest point
- The deepest the running total fell below the starting point at any moment. It shows how much had to be absorbed along the way, even in sessions that finished ahead.
- Longest losing run
- The longest stretch of consecutive losing calls in the session.
Why the sessions disagree
The standard deviation of one call is 0.542 units, so over 120 calls it is about 5.94 units. The expected net of -1.20 is small next to that. Sessions several units up or down are normal, and a single session says almost nothing about the edge.
What a much longer run shows
Running the same generator, seed 7741, for 200,000 calls gives an average result of -1.131% per unit. The exact value is -1.000%. The two are close but not identical: even 200,000 calls leave a sampling error of roughly 0.121%.
Limits of a simulation
- It is a model run on a computer, not a record of anyone's real results.
- A different seed gives different sessions; the seed only makes this run repeatable.
- It uses an ordinary pseudo-random generator, not the hash construction that verifiable services publish (see a worked hash check).
- The exact formulas, not the simulation, are the authority on the expected value.