This table shows the playing strength (PR) of the best iOS backgammon apps. I’ve asked the three best backgammon players on this planet to spend hours analyzing matches (who all happen to be PC programs). Here’s their result. Details on how I get the numbers below.
| App | total match length | PR |
|---|---|---|
| XG Mobile | 46 | 0.5 |
| True Backgammon HD | 40 | 1.1 |
| BGNJ | 40 | 2.9 |
It seems that XG Mobile is leading the pack by a margin. And all 3 PC programs agree on this
- XG2 ranks XG mobile at 0.6, True BG at 1.2 and BGNJ at 3.2
- BGBlitz ranks XG mobile at 0.5, True BG at 1.0 and BGNJ at 2.5
- Gnu BG ranks XG mobile at 0.4, True BG at 1,2 and BGNJ at 2,9
Again for comparison: no human being is known to have a PR below 2.4 – these apps play a mighty good game of backgammon.
The “total match length” is the total number of points of all matches I’ve fed into the programs – 40 means e.g. eight 5pt matches.
Now, how did I get these numbers, or “the daring quest to figure out which is really the strongest backgammon app for mobile devices, without playing anywhere strong enough to have a valid opinion.”
As you can see in the mighty list, it looks like XG Mobile plays a tad stronger than its two main competitors for the crown of strongest backgammon app – True Backgammon HD and Backgammon NJ. But, you may ask, isn’t it a dumb idea to use XG Mobile’s big brother – XG2 for the PC – to determine whether this app or that app that play nearly as well as XG2 makes a tiny mistake? And you are right to ask this. XG2 might be biased, using a similar AI.
So this is how I get an unbiased rating, without being able to judge these apps myself, at all.
- I select all backgammon apps I know that achieve a superhuman PR of less than 2.4 in my review. Currently these are the three apps in the table.
- I’m playing a couple of matches to five and seven points against these apps, all on their highest level.
- I’m feeding the match files I get into
- XG2 for the PC, analyzing on Roller++ level (slow)
- BGBlitz for the PC, analyzing on 5-ply level (slooooow)
- Gnu Backgammon for the PC, analyzing on 4-ply level (slooooow, but 5-ply would take days)
- I find out what total lost equity the three superhuman PC programs find in their hour-long analysis sessions.
- I show you the results in a big table (see below)
- I calculate the average PR from the average equity loss across the three programs.
- I show you THE ULTIMATE CHAMP (see above).
The table with all the data
If you want to find out all the boring details, here’s an image of the excel table I was using, along with some explanations.
The upper table contains a listing of all matches I’ve analyzed. XG-Champion is XG2, True BG BGBlitz2 is True Backgammon, and BGNJ is, well, BGNJ.
The columns labelled XG contain XG’s PR calculation, and the equity loss it determined – overall, then checker play, then cube. “eq loss dice” and “eq loss cube” add up to “equity loss”, obviously.
The next columns show the analysis result of BGBlitz and of Gnu Backgammon. Instead of BGBlitz’s own PR rating I’m calculating its PR rating from its equity loss and XG2’s PR rating(*). I’m doing the same for Gnu BG which doesn’t know the concept of BG.
The final column, overall PR, is the average PR rating from XG2, BGBlitz and Gnu. I then calculate the average PR of each app as the average of all matches (NOT weighing a 11pt match higher than a 5 pt match, which is a mistake that I’ll correct some day but which doesn’t affect the overall outcome).
The second table with the green column shows the result of the whole thing – the PR of each app over the matches in the upper table. The next columns show how many points were played to overall, and what XG2, BGBlitz and Gnu individually conclude. The detail values are different, but all three rank the three apps consistently.
(*) unexpectedly, the PR calculation of XG2 and BGBlitz is not the same – the same equity loss in the same game doesn’t yield the same PR. Probably they have slightly different opinions about what a decision is. I decided to consistently follow XG2’s PR approach and calculated e.g. the BGBlitz PR from the XG2 PR, the XG2 equity loss and the BGBlitz equity loss.
ps. if you really look closely at the table you’ll notice a mysterious row in the middle. In one match, BGBlitz calculated XG Mobile’s PR as -0.1 – I don’t know why it did it, but it seems that XG mobile played better than what BGBlitz thought is the perfect move?
Comments are highly welcome!