Little Golem#1 · Alan Hensel
6 Nov 2026
Twixt has the "pie rule", allowing a player to swap the first move if it is too strong. It is not obvious, however, especially to a beginner, what a strong or weak first move looks like. It might not even be immediately obvious that the pie rule is necessary, even to Alex Randolph, who originally had the game published without it.
But I have data. Twixt Commentator caches all completed games that people visit. It also caches completed games that web crawlers visit, and they are very thorough. I've also recently added cron jobs that make it even more thorough. I have 99.9+% of the 105,000 Twixt games stored in the Commentator database. I can do some statistics.
Here's a number of pegs histogram: https://twixt-commentator.duckdns.org/stats/pegs
Notice the nice smooth curves, peaking around 24 or 25 pegs. (Having 100,000 games will do that.) Notice also that there are actually 2 curves, overlapping, making the graph look a bit like a stegosaurus.
What that is telling us is that Twixt games played on Little Golem have been more likely to end with an odd number of pegs.
So why is that? Notice, this is number of pegs, not number of moves, so "swap" plays no part in this statistic. I think this is saying is: 1) players on Little Golem, many of whom are or have been beginners while playing on Little Golem, were more likely to choose a move that is too strong than a move that is too weak or balanced, and 2) the first move really does matter.
But then what's a good first move? I did some more statistics. In particular, I chose "Elo-adjusted expected-score residuals", which has been used in Chess for evaluating best first Chess moves. In part, I chose this statistic so you can google it if you really want to understand it deeply. But the tl;dr is that in the resulting table, pink is good, blue is bad:
https://twixt-commentator.duckdns.org/stats/first-moves
One thing I find interesting about this is that even back in 2003 when Twixt started on LG, good players had settled on d3 as a good neutral first spot, based on their intuition alone, and these statistics back it up. The statistics also suggest a range of other good first moves.
Just to get a second opinion, I manually collected data from an AI (Twixtbot), picking the neural net's evaluation of the strength of each of the 132 unique first moves, put them in a spreadsheet, and had Gemini color the spots accordingly. This color scheme is different, so for this one, blue=weak, purple=neutral, red=strong.
https://docs.google.com/spreadsheets/d/1TZdIew7sPJVvKPtMZb4o6MLTh2QcB94NZpMe7AcBTE4/edit?gid=0#gid=0
The general shape is the same. There are minor differences, which is a good reminder not to stare at these statistics too closely. Like the early Twixt masters, please engage your intuition, and have fun!