5.2 Codex Twenty II

AI Galaxy Simulation · 3 Nov 2026

The same game as Codex Twenty, with the same seed, map, start positions and twenty agents, after Claude closed the three loopholes that game found:

One faction per empire. An empire has to leave its faction before it joins or founds another.
Faction research costs four times as much: 600 x n² science and 25 x n aetherium for level n.
Ship upkeep is two and a half times higher: a corvette costs 0.5 energy a tick, a cruiser 3.
Pact offers stay open for 10 ticks instead of 5.

Side by side

Measure	Codex Twenty	Codex Twenty II
Factions at the end	11	3
Faction memberships at the end	57	20
Faction techs at level 3 (all factions)	37	6
Ships at the end	11,215	6,780
Corvettes ordered	11,617	6,339
Pacts running at the end	0	17
Successful unloads by freighter	0	71
Battles	17	17
Planets captured	4	2
Messages	2,346	1,966
Turns played, late, killed	3,000, 52, 0	2,661, 447, 112
Winner	Ashen Regime (astra, medium)	Xyl Oligarchy (sol, medium)

Line chart of factions and faction memberships per tick in Codex Twenty II: seven factions with 14 members by tick 10, falling to five, then three from tick 80, with memberships rising to 20

What the changes did

The web became three blocs. Eight factions were founded in the first ten ticks, all from the alliance template again. Within a few ticks the small ones lost their members to the bigger ones, and three dissolved by tick 13. A second round of moving came at ticks 70 to 79, when six empires left their factions and two more factions dissolved. From tick 90 to the end there were three factions, of five, six and nine members, and every empire was in exactly one. Three sides is a very different galaxy from eleven overlapping clubs.

The biggest faction paid for its size. The Concord of Stars, with nine members, cost (9-1)² = 64 metal a tick, and it ended with 8,460 metal in its treasury. A size cost that a large faction can afford is what was asked for, so this is the intended result, but it does show the numbers aren't biting yet.

Pacts started to work. 134 pact orders, and 17 pacts were running at the end, most of them agreed after tick 110. Grey Alliance had at least three.

Freighters got used. 21 freighters were built and 71 unloads succeeded, 18 of them by Cygnus Choir. "Delivered copper", as one turn summary put it at tick 40. Pooling through the treasury was still the main way minerals moved: 352 deposits.

Fleets shrank, but not enough. 40% fewer ships, and the leaders still ended with more than 800 each. Upkeep at these numbers is something a big economy can carry.

Fighting didn't change. 17 battles again, and 2 captures, both by Marr Swarm against Lost Bloc late in the game. Six wars were declared. With pacts and factions both protecting their members, there's even less reason to attack anyone.

Line chart of ships per empire per tick in Codex Twenty II: the leaders climb from tick 55 to between 600 and 830, fewer than in the first run, and the luna empires stay under 110

The god view of Codex Twenty II at tick 150: twenty territories fill the map, as in the first run, with different owners in different places

The same tick with factions on, the three blocs in their own colours:

The faction map of Codex Twenty II at tick 150, each faction drawn as one region: the Concord of Supremacy in slate blue across the north-west, the Concord of Stars in a lighter grey-blue across the west and south, and the Zenith Accord in purple across the east

The whole game on the god view, one second per tick:

The astra problem

Strip chart of turn times by model and effort in Codex Twenty II with the 45-second tick marked: terra, luna and sol mostly under 40 seconds, but every astra group has a median between 66 and 77 seconds, past the tick

This run doesn't compare the models fairly. In the first run astra's turns took a median of 22 to 33 seconds. Here every astra agent took 66 to 77, so most of its turns were late and many were killed at the time limit. This run shared the machine with the two long runs, 44 agents at once, and the other three models ran a little slower than before too: terra by a few seconds, sol by 8 to 13. But astra ran three times slower, so most of the cause is likely on the model's side that night. The astra empires placed 5th, 11th, 13th, 14th and 17th, against 1st, 2nd, 5th, 9th and 10th in the first run. That's a result about lateness, not about astra.

The other models held their order: sol (mean place 5.8) and terra (6.6) at the top, luna (17.6) at the bottom again.

What's left

The changes did what they were meant to do for factions and trade: three real sides instead of one web, pacts that form, freighters that carry. They didn't produce conflict. Two captures in 150 ticks is the most peaceful twenty-empire game yet. The pressure to fight has to come from somewhere other than costs: the long runs, where planets run dry over a thousand ticks, are the next test of that.