A living research notebook on processes, model identification, emergence, and viability.
No boid knows the flock. Local rules produce a global trace.
Most things do not arrive with their causes attached. We hear a sound, watch a flock, read a model's answer, or encounter an institution through its decisions. What we receive are traces of processes we cannot see in full.
This repository studies two bounded questions:
- Model identification: What can a finite observer identify about a process from partial traces, and what changes when observation is joined by construction, intervention, and revision?
- Viability: What constraint architecture lets an optimizing system remain correctable and viable as its capabilities grow?
The shared foundation is deliberately modest: typed stochastic processes, composition, observation, and prediction. Identity is relative to declared tests. Learning and intelligence require tasks, losses, and resource bounds. Phenomenal consciousness is not derived.
This is a research notebook, not a theory of everything. It contains established mathematics, controlled toy experiments, working hypotheses, simulations, essays, and fiction; their evidential status is kept separate.
| If you want to… | Start here |
|---|---|
| understand the project without learning its internal vocabulary | Start Here: From Traces to Viable Intelligence |
| explore the complete live notebook | Read the online notebook |
| change rules and watch a system evolve | Run the Web Emergence Explorer |
| inspect the mathematical foundation | Read the Foundations Reconstruction |
| see the measured model-identification results | Run the Inverse-Reconstruction Benchmark |
| examine whether capability belongs to a controller or a complete situated stack | Read The Agent Is Not Where the Model Ends |
| distinguish generated, exposed, selected, and observer-attributed competence | Read Competence, Constraint, and Verification |
| read the current formal synthesis | Open The Viable Corridor |
| browse unfinished questions before they become claims | Enter Ideas |
| Layer | Role | Main artifacts |
|---|---|---|
| Foundation | Defines the process language and marks what does not follow from it. | Foundations Reconstruction · What This Project Does NOT Claim |
| Model identification | Studies how traces constrain typed generators or candidate process models and how interventions change the evidence. | The Generator Question · From Trace to World-Binding · Inverse-Reconstruction Benchmark |
| Viability | Studies optimization under explicit dynamical and substrate constraints. | Optimization and Its Blindness · The Viable Corridor |
The Generator Question remains active as a typed construction and model-identification programme. What the foundations audit rejects is the unqualified generator as a mathematical primitive, universal law, or complete explanation. Current uses must say what generates what: a transition kernel, developmental process, grammar, proposal mechanism, or another declared map. The earlier universal claims remain visible as research history. Identity, consciousness, cooperative intelligence, culture, and the narrative work remain active layers around this spine, each with its own scope and status. The Conceptual Map shows how they connect.
The situated-competence extension does not replace that question or add a new primitive. It asks which capacities a typed process can express under changed constraints, embodiment, selection, interaction history, and verification. Its central synthesis is Competence, Constraint, and Verification.
The Web Emergence Explorer is a zero-install cellular automaton. Change the initial pattern or draw directly into the world while Shannon entropy and spatial mutual information update in real time.
git clone https://github.com/frnkptrln/systems-and-intelligence.git
cd systems-and-intelligence
pip install -r requirements.txtFor development and documentation work, install the matching environment:
pip install -r requirements-dev.txt # core runtime plus pytest
pip install -r requirements-docs.txt # core runtime plus MkDocs and PDF toolsRun the animated Boids model shown above:
python simulation-models/emergent-dynamics/boids-flocking/boids.pyRun the compact inverse-reconstruction benchmark:
cd lab/benchmarks/inverse-reconstruction
python inverse_benchmark.pyMore executable work:
- Simulation → Theory Map — where the simulations touch the claims and open questions
- Lenia — continuous cellular automata
- Reaction–Diffusion — pattern formation from local chemistry
- Nested Emergence — coupled dynamics across scales
- Referee Benchmark — when does a self-revising loop actually improve, exactly measured
- Situated-Stack Benchmark — one controller under changed sensors, actuators, bodies, environments, goals, and observer lenses
- Constraint-Release Benchmark — separates behavioral exposure from a lens-only score increase under a fixed seeded substrate
- Agentic Identity Suite — experimental tests for persistence, binding, and availability
| Path | Purpose |
|---|---|
ideas/ |
Small observations and open questions before classification or synthesis |
theory/ |
Formal and semi-formal arguments, maps, and reference material |
lab/ |
Benchmarks, metrics, reusable tools, and experimental infrastructure |
simulation-models/ |
Executable toy systems grouped by research theme |
papers/ |
Tighter working drafts intended for external criticism |
book/ |
A longer, evolving reading path |
fiction/ and logs/ |
Narrative stress tests and applied architecture notes, not evidence |
meta/ |
Project architecture, epistemic rules, registries, and open problems |
The live notebook carries the full navigation. The repository's information architecture explains how new material moves between these layers.
- The foundation is an audit and dependency map built from established mathematics, not a claim of novel mathematics.
- The inverse-reconstruction results are measurements in small, controlled systems.
- The Viable Corridor is a conditional model: necessity is argued under its assumptions, sufficiency remains conjectural, and broader social mappings are heuristic.
- Functional architectures can be tested; subjective experience is not inferred from behavior or organization.
If a page sounds stronger than these boundaries allow, What This Project Does NOT Claim controls the reading.
This project is maintained by Frank Peterlein as an independent research notebook developed in sustained collaboration with AI systems.
Why it exists, in one honest sentence: curiosity — some questions do not let go.
Corrections, counterexamples, critical review, and reproducible experiments are welcome. Licensed under the MIT License.