Experiments

Archive

The experiment code is organized around study records. A study is the scientific unit: title, status, seeds, result location, source, key finding. The shared substrate code is a library.

  • experiments/studies/: one directory per study, 009_intrinsic_predictive_gradient, 012_nonlinear_mlp_prediction_head, and so on, each with its experiment.json, README, source, and results.
  • experiments/lib/: the Lenia and protocell substrate lineage and the affect measurement utilities, installed flat so historical imports still work.
  • experiments/registry.json: the canonical index, validated by experiments/scripts/validate_registry.py.

Read the sections below as a chain. ask what a cellular automaton can do without agency. asks what changes once action-observation loops exist. isolate the prediction architecture. test whether social targets, direct selection, contrastive losses, or communication change the answer. They do not. The nulls are load-bearing.

Research Roadmapfive priorities for the empirical program1Universal GeometryConfirm affect structure in uncontaminated systemsV1, V3-V6, VLM●2Integration DynamicsWhat architectural features produce Φ increase under stress?V8-V12, V14-V17●3Individual PlasticityWithin-lifetime adaptation bridging the attention bottleneckV18, V19◐4Social-Scale ΦMeasure collective integration in multi-agent systemsplanned○5Developmental OrderingTest emergence ladder predictions in human developmentproposed○●complete◐active○planned