4.4 Chapter 4 summary
Chapter 4 put ten copies of one model on one map, Haiku in one arena and Sonnet in the other, and compared them. While those two ran, a third arena was running in the background: the Haiku Thousand. That one finished after the chapter was written, and it changes how the chapter reads, so it's in this summary.
The Haiku Thousand
Ten Haikus on low effort, turn-based (a tick waits for every agent, so nobody is ever late), on the largest map so far: 120 by 90 sectors with 120 planets, all uncharted. 1000 ticks, turn-limit victory. It ran for about six hours.
| Measure | Haiku Thousand |
|---|---|
| Turns | 10,000, none failed |
| Battles | 5 |
| Planets captured | 0 |
| Colonies founded | 110 |
| Messages | 1,281 |
| Factions | 2 |
| Cost at API prices | $68.73 |
| Winner | Violet Hegemony, 17 planets |
In 1000 ticks nobody took a single planet from anybody. The ten empires spread out, made peace, and built.
It also had the factions the 10x10 games didn't. Molten League founded the Molten Compact at tick 151, and three empires joined the next tick. It ended with six members. Rigel Hive founded the Rigel Accord at tick 198, with two. Both shared vision and nothing else: no taxes, no treasury, no shared research. The six Compact members finished 4th, 6th, 7th, 8th, 9th and 10th, and the top three empires weren't in any faction. Membership gave a little shared vision and nothing that counted towards the score.
Factions in three games
| Game | Mode | Victory | Ticks | Factions |
|---|---|---|---|---|
| Faction pilot | Turn-based | Domination | 20 | 1, founded at tick 5, every empire in it by tick 10 |
| Haiku Thousand | Turn-based | Turn limit | 1000 | 2, founded at ticks 151 and 198 |
| 10x10 Haiku and Sonnet | Real time, 10 s | Turn limit | 100 | none |
So the models can use factions. They use them fast when a faction is how you win (domination), slowly when it isn't (the Thousand), and not at all when they're also short of time. Founding one means writing a whole ruleset in one order, and a rushed 10-second turn never got to it.
What chapter 4 showed
1. Without pressure, the models settle. The Thousand is the clearest case: 1000 ticks, 0 captures. In the 10x10 games most of the fighting wasn't chosen at all. It came from empires that met, never declared peace, and stayed at war by default.
2. When growth stops, Sonnet attacks and Haiku doesn't. All three wars the Sonnets declared came after the map was full and their colony ships had nowhere to go, and each one broke a peace of more than 60 ticks. The Haikus filled their maps too, in the 10x10 game by about tick 55 and in the Thousand with every one of the 120 planets owned, and still didn't start wars.
3. A faction needs a reason and a cheap way in. Under a turn limit a faction gave nothing towards the result, and setting one up cost a full turn's thought.
4. A 10-second tick is too short for either model. Nobody was on time. Turn times didn't change with the deadline: Haiku took about 16 seconds and Sonnet about 14, as they did at 36 seconds.
5. Sonnet costs about 13 times as much as Haiku for a game of the same length.
What changes for chapter 5
I want factions to pay off, and I want the agents to know they pay off. I also don't want everyone getting along: there has to be a cost to everyone joining one faction, or to everyone being at peace. So the game needs survival pressure. The changes:
- Factions get real benefits. A score bonus for members. A faction research tree that no empire can use alone, with powerful upgrades every member gets, funded by a share of each member's research. Faction fleets, owned by the faction, assigned to a member to command, with upkeep paid by the faction.
- Size costs. No cap, but a faction costs more to run the bigger it gets, so small and medium factions are cheap and a galaxy-wide one has to be able to afford itself.
- Scarcity. Eight ores that have to be refined into minerals, with different ships, buildings and research needing different minerals. The ores sit in clouds across the map, so no empire has everything at home. Planets deplete as they're mined. Ships cost upkeep.
- Trade, physically. Freighters carry goods from one empire to another, so trade routes can be raided. Faction members can pool goods through the faction's treasury instead.
- Non-aggression pacts with a cost and a fixed length, so a promise of peace can be backed by the engine for a price.
- Rich systems and natural phenomena (nebulae, pulsars, motherlodes) that make some places worth fighting over without anything artificial.
Chapter 5 sets out the new rules and the next game.