6 The Next Generation
Chapter 5 ended with a plan for this chapter: points for every capture and every battle won, a bonus for your share of all planets and ships, a rank in every brief, and a knockout of three arenas, twenty empires, then ten, then five. Claude built all of it. It hasn't been run.
Before running it, I wanted to think about the corvettes. In Codex Marathon the agents sent 10,809 corvette build orders, 2 for cruisers and none for frigates. Working out how to make bigger ships worth building turned into a rebuild of most of the game. I chose not to run the knockout on the old rules, so version 1 ends with chapter 5, after sixteen games, and this chapter starts version 2.
- 6.1 What we've learned so far looks back over the sixteen games: what they showed about the models, and what's wrong with the game they played.
- 6.2 Version 2 changes sets out the new game in full.
Version 1 and version 2
| Version 1 | Version 2 | |
|---|---|---|
| Map | Planets on a grid, with ore clouds | Systems of stars and bodies, in five galaxy layouts |
| Resources | Deposits in the clouds, and later in waves | Deposits on every body, some rich; belts and gas giants mined by ships |
| Goods | One stockpile for the whole empire | Held at each body, moved by ship |
| Economy | Buildings with levels, refining across the whole empire | Buildings with inputs, outputs and workers; food; four kinds of refinery; credits and tax |
| Research | Levels in a handful of fields | A tree of 91 nodes, with doctrines |
| Ships | Fixed classes | Designs, from 6 hulls and 35 components |
| Combat | One shot per ship per round | An opening salvo, shots per mount, shields, armour and point defence; only frigates and larger can besiege a planet |
| Capture | Takes the planet | Takes the planet and everything stored on it, faction depots included; plunder without capture |
| Mysteries | None | Anomalies, wormholes, guardians and signal chains |
The core stays the same: ticks, fog of war, fleets and their postures, messages, stances, pacts, factions, the score and the rank, and the plain HTTP API the agents play through.
What I hope to see
- Bigger ships. Fleets of frigates, cruisers and battleships instead of ten thousand corvettes, with the odd capital ship as a big project.
- Specialists. Empires that run one or two refineries and trade for the rest.
- Supply lines. Mining fleets on the belts, freighters on routes, and raiders hunting both.
- Reasons for war. Planets taken for what's stored on them, and depots raided.
- Mysteries found and shared.
And two open questions: does anything still move after the land grab, and can an agent run an economy this deep inside a 45-second turn?
Method
The first game is the V2 Pilot: eight Codex agents, terra and luna at low and medium effort, two of each, in a two-arm spiral of 56 by 42 sectors. It runs in real time, 120 ticks of 45 seconds, with a turn limit. It's a test of the new economy: local production, ship designs, mining and supply routes.
After that comes the knockout, the Codex Gauntlet, on version 2 rules. Twenty Codex agents, five each of sol, astra, terra and luna at mixed effort, play three rounds in real time: twenty empires, then ten, then five, each round on a new seed and a smaller two-arm spiral (90 by 68, then 64 by 48, then 45 by 34), 150 ticks of 45 seconds a round. Every agent is told it must finish in the top half to go through, and in the final it must win. Each one keeps an identity paragraph that it rewrites as it plays, and at the end of a round it writes a handover for the next one.
The data is the same as before: every turn with its timing, cost and the agent's own summary, every order sent, and the god log.
Caveats
Everything changed at once again, so a difference from version 1 can't be pinned on any one change. No agent has played version 2 yet: its numbers were tuned against a battle simulator and production checks, not against real play. And the pilot is small and short, eight empires for 120 ticks. It can show whether agents can work the new economy, not how a long game plays.