The research archive · Volume I
No polished mythology. A public record of hypotheses, methods, unexpected outcomes and limitations — kept for one artificial society.
The sequence (P numbers are the project’s internal order)
Introduction · read this first
How the artificial society works; what tick, world, seed, arm and fork mean; how experiments are designed and analysed; and what they can and cannot show.
Read the paper →21 min read
P10–P12 · Autonomous care
Switching off the engine's automatic household food sharing raised child starvation by 1.425 deaths per world; residents' own care compensated only in part.
+1.425more child starvation deaths per world once the engine stopped sharing food within families (P10; rose in 31 of 40 paired worlds)
Read the paper →Notebook entry ↗20 min read
P17–P18 · Tool economics
Under scarcity a free tool and a food reserve had large positive value and a free house did not; making a tool from scratch was not shown to pay for the learning controller mind9.
+29.5net alive person-ticks from making a tool from scratch under scarcity — not shown to pay (95% CI −18.4 to 77.8)
Read the paper →Notebook entry ↗18 min read
P19 · Language-model residents
In the food-constrained world, residents controlled by DeepSeek-V4.1-Flash-1017 had lower welfare than both baseline controllers in all 30 worlds; private memory changed plans but not welfare.
30 / 30food-constrained worlds in which DeepSeek-V4.1-Flash-1017 residents had lower welfare than both baseline controllers
Read the paper →Notebook entry ↗18 min read
P14 · Bounded beliefs
A bounded belief predictor forecast a child's energy better at +1 and +4 ticks but failed the preregistered +12-tick gate, so no action experiment followed.
−0.63%world-relative forecast improvement at +12 ticks — short of the preregistered 5% gate, so the study stopped
Read the paper →Notebook entry ↗17 min read
Field notebook · Vol. I
The same four experiments as an observer’s hand-annotated journal — turn the pages, follow the sketches, then step into the papers.
Field notes & essays
Short essays on method and direction — part of the same open record.
Parallel worlds are a future research direction: identical starting conditions, different interventions, and careful comparison.
Research questionSimulation onlyThe point of a research archive is not to prove every idea right, but to make the reasoning visible.
Research questionSimulation only