Eco in Space - A Total Conversion Mod #17

New issue

Open

opened 2026-08-20 03:15:11 +00:00 by coilyco-ops · 0 comments

coilyco-ops commented

2026-08-20 03:15:11 +00:00

Member

Conclusion

Total convert Eco into a space setting, incrementally, one slice at a time. Not as a themed reskin, but because the setting is what grants social license for non-human first-class entities. That license is the actual deliverable. Space is the mechanism.

How this got here

This started as "pick a game for the evening" and walked through most of the space-sim genre before arriving somewhere else entirely. Recording the chain because the reasoning is the useful part, not the destination.

Pass 1, what to play. Slow space game, conceptually infinite, single persistent world, not X4. Landed on ΔV: Rings of Saturn, which was correct for the mood and got bought and downloaded. Not the point of the evening in the end.

Pass 2, the Empyrion wall. Empyrion is the one game that sits between the engineering-first wing (Space Engineers, Stationeers, From The Depths) and the world-first wing (No Man's Sky, Starfield, Astroneer). Runs consistently die around 150h at the same place: the self-sufficient fortress CV never becomes a combat CV.

Diagnosed as three RE2 rules rather than a skill gap:

Planetary POIs versus CVs is a designed 4x disadvantage - CV weapons do half damage to planetary POI shields, planetary POIs do double damage to CV shields.
Shield penetration exists - armor is mandatory, and a volume-optimized fortress is thin-skinned by construction.
Combat is mostly piloting - community estimate is 70 to 90 percent flying, not ship. Orbit one side, disarm that side, kill generators.

The documented late game is the carrier: one hull with farming, production, storage, repair, and docking faces, launching specialized craft. The fortress never becomes the warship. It becomes the thing warships launch from. Also established that CV-to-CV docking works but does not nest, that the home CV should never land, that the CV T1 telepad does same-star transport, and that furnaces cannot be placed on a CV at all.

Pass 3, the pivot. Attention was on system design, not play. Reframed the question from "what to play" to "what to mod."

Pass 4, mod filter. No Lua, no Harmony, no BepInEx, no injection frameworks. First-party sanctioned APIs only. Java allowed with penalty. Ten candidates surveyed. Stardeus won on friction and first-party C#. Eco was already in the top tier on the strength of the official ModKit, and Kai has a published Eco modding writeup and an existing MCP over the game state.

Pass 5, the real goal. The mod is an LLM running a space empire. Filter changed again to games whose mod layer can hold a socket open. Space Engineers died here, since programmable blocks cannot make HTTP requests and the whitelist blocks network libraries. Stellaris died here, being script-only with no sanctioned external control.

Pass 6, the axis that mattered. Not empire size. Whether the agent is an optimizer or an inhabitant. Starsector and X4 both cast it as an optimizer that makes number go up, which is subtraction from games already played for hundreds of hours. Stardeus wins that axis natively, since the game already casts the player as the ship's AI.

Eco's problem was stated as having no native entity to drop an agent into. An agent in a human seat would be resented, correctly. Which is exactly what the space conversion fixes: in a space mining setting, an autonomous machine is unremarkable and needs no argument.

Why this over the alternative

The alternative was DeepSeek running Stardeus. It is cheaper, diegetically perfect, and finished the moment it works. Joy from a demo decays.

Eco has 2,660 hours of accumulated fluency behind it, roughly 50x anything else in contention, plus an existing mods repo, an MCP over game state, and a community already in the room. Every long-run engagement pattern in the library points the same way: persistent worlds with other people in them are the only thing that holds attention past 150 hours.

Sequencing

Total conversions kill people. Do not build the conversion in order to get the robot.

Build the smallest space-flavored slice that makes one robot legitimate as a first-class entity.
Put it in front of the community and see whether anyone plays it.
If they do, there is a year of work with an audience. If they do not, that was learned in a month rather than nine.

Agents come later and natively. A robot in a robot seat is fine where a robot in a human seat is not.

Adjacent, not blocking

Merchant agent - viable, but as a headless market management script rather than an entity. The MCP already covers market, stores, trades, find_trade, fair_price, and trade_watchers.
DeepSeek runs Stardeus - worth two evenings as a prototype, purely to learn what an LLM can actually hold in a colony sim. Load-bearing for later agent work here. Not the project.

Rejected

Starsector and X4 - optimizer-shaped agents, and X4 additionally has to beat its own incumbent automation.
Elected-official agent in vanilla Eco - the community would resent an AI overlord, and rightly.
Eco total conversion as a first move - correct destination, wrong first step.

## Conclusion Total convert Eco into a space setting, incrementally, one slice at a time. Not as a themed reskin, but because the setting is what grants social license for non-human first-class entities. That license is the actual deliverable. Space is the mechanism. ## How this got here This started as "pick a game for the evening" and walked through most of the space-sim genre before arriving somewhere else entirely. Recording the chain because the reasoning is the useful part, not the destination. **Pass 1, what to play.** Slow space game, conceptually infinite, single persistent world, not X4. Landed on ΔV: Rings of Saturn, which was correct for the mood and got bought and downloaded. Not the point of the evening in the end. **Pass 2, the Empyrion wall.** Empyrion is the one game that sits between the engineering-first wing (Space Engineers, Stationeers, From The Depths) and the world-first wing (No Man's Sky, Starfield, Astroneer). Runs consistently die around 150h at the same place: the self-sufficient fortress CV never becomes a combat CV. Diagnosed as three RE2 rules rather than a skill gap: * **Planetary POIs versus CVs is a designed 4x disadvantage** - CV weapons do half damage to planetary POI shields, planetary POIs do double damage to CV shields. * **Shield penetration exists** - armor is mandatory, and a volume-optimized fortress is thin-skinned by construction. * **Combat is mostly piloting** - community estimate is 70 to 90 percent flying, not ship. Orbit one side, disarm that side, kill generators. The documented late game is the carrier: one hull with farming, production, storage, repair, and docking faces, launching specialized craft. The fortress never becomes the warship. It becomes the thing warships launch from. Also established that CV-to-CV docking works but does not nest, that the home CV should never land, that the CV T1 telepad does same-star transport, and that furnaces cannot be placed on a CV at all. **Pass 3, the pivot.** Attention was on system design, not play. Reframed the question from "what to play" to "what to mod." **Pass 4, mod filter.** No Lua, no Harmony, no BepInEx, no injection frameworks. First-party sanctioned APIs only. Java allowed with penalty. Ten candidates surveyed. Stardeus won on friction and first-party C#. Eco was already in the top tier on the strength of the official ModKit, and Kai has a published Eco modding writeup and an existing MCP over the game state. **Pass 5, the real goal.** The mod is an LLM running a space empire. Filter changed again to games whose mod layer can hold a socket open. Space Engineers died here, since programmable blocks cannot make HTTP requests and the whitelist blocks network libraries. Stellaris died here, being script-only with no sanctioned external control. **Pass 6, the axis that mattered.** Not empire size. Whether the agent is an **optimizer** or an **inhabitant**. Starsector and X4 both cast it as an optimizer that makes number go up, which is subtraction from games already played for hundreds of hours. Stardeus wins that axis natively, since the game already casts the player as the ship's AI. Eco's problem was stated as having no native entity to drop an agent into. An agent in a human seat would be resented, correctly. Which is exactly what the space conversion fixes: in a space mining setting, an autonomous machine is unremarkable and needs no argument. ## Why this over the alternative The alternative was DeepSeek running Stardeus. It is cheaper, diegetically perfect, and finished the moment it works. Joy from a demo decays. Eco has 2,660 hours of accumulated fluency behind it, roughly 50x anything else in contention, plus an existing mods repo, an MCP over game state, and a community already in the room. Every long-run engagement pattern in the library points the same way: persistent worlds with other people in them are the only thing that holds attention past 150 hours. ## Sequencing Total conversions kill people. Do not build the conversion in order to get the robot. 1. Build the smallest space-flavored slice that makes one robot legitimate as a first-class entity. 2. Put it in front of the community and see whether anyone plays it. 3. If they do, there is a year of work with an audience. If they do not, that was learned in a month rather than nine. Agents come later and natively. A robot in a robot seat is fine where a robot in a human seat is not. ## Adjacent, not blocking * **Merchant agent** - viable, but as a headless market management script rather than an entity. The MCP already covers market, stores, trades, find_trade, fair_price, and trade_watchers. * **DeepSeek runs Stardeus** - worth two evenings as a prototype, purely to learn what an LLM can actually hold in a colony sim. Load-bearing for later agent work here. Not the project. ## Rejected * **Starsector and X4** - optimizer-shaped agents, and X4 additionally has to beat its own incumbent automation. * **Elected-official agent in vanilla Eco** - the community would resent an AI overlord, and rightly. * **Eco total conversion as a first move** - correct destination, wrong first step.