Hi-lo guide
Bamboo Hi-lo: a 4 and the spread of results
Bamboo labels this page; it does not point to a special version. The subject is the face-up 4, and in particular how much results scatter when the same call is made again and again.
The basic numbers for a 4
With ace low and king high, a 4 has three ranks below it (A, 2, 3), nine above it (5 to K) and one equal rank. In this model the next card is uniform over the 13 ranks, so P(higher) = 69.23% and P(lower) = 23.08%. The remaining 7.69% is the tie, which loses under the strict rule.
This page uses a low edge of 0.5% on purpose. Higher then pays 1.4372x and lower pays 4.3117x. The expected loss per unit is only 0.50%, so it is easy to see how small the average is next to the natural swing.
One call: mean and standard deviation
For a single call the mean result is -0.0050 units on either side. The standard deviation is 0.663 units for higher and 1.817 units for lower. In words: the typical distance from the mean on one call is more than a hundred times the size of the mean itself.
Many calls: higher from a 4
| Rounds | Expected net | SD of net | Rough 95% band (normal approx.) |
|---|---|---|---|
| 25 | -0.13 | 3.32 | -6.76 to 6.51 |
| 250 | -1.25 | 10.49 | -22.23 to 19.73 |
| 2,500 | -12.50 | 33.17 | -78.83 to 53.83 |
The expected net grows in step with the number of calls, but the standard deviation grows only with its square root. After 25 calls the band is wide and centred close to zero, so either a gain or a loss is common. After 2,500 calls the expected loss of 12.5 units is still smaller than one standard deviation of 33.2. At a 0.5% edge the average needs a very long run to stand out from the noise.
Many calls: lower from a 4
| Rounds | Expected net | SD of net | Rough 95% band (normal approx.) |
|---|---|---|---|
| 25 | -0.12 | 9.08 | -18.29 to 18.04 |
| 250 | -1.25 | 28.72 | -58.70 to 56.20 |
| 2,500 | -12.50 | 90.83 | -194.16 to 169.16 |
Same expected net, roughly double the spread. A session of lower calls from a 4 can finish well ahead or well behind, and the result says very little about the game. Neither table means the edge goes away. It is always there; it is just quiet next to the variance at these sample sizes.
What this means for reading your own results
- A winning week of hi-lo calls is not evidence of skill or of a generous game. The spread above allows it easily.
- A losing week is not evidence of a rigged game either. Checking a service needs hash verification, as in the hash fairness guide, not results.
- The expected loss is real and adds up. Over the 2,500 calls above it is 12.5 units at this low edge, and five times that at a 2.5% edge.
Other ranks are covered in the hi-lo overview.