How You Live in Your Own Simulation of Reality — and How We Can Still Share a World
A Theory of Persistence, Emergence, Learning, and Correctable Intelligence
Evolution by Emergence v23.0 — a substrate-agnostic candidate theory of persistence and cumulative organization, with a machine-checked specialization for correctable intelligent systems. New here? Start with CORE.md: what a functional world model is, then one model of learning, read as a network, and what follows from it, in eight steps on one page, checked in one Lean file.
Evolution by Emergence (EbE) is a theory of persistence. It starts from learning in a neural network, read as a network. A configuration of connections exists only while it pays its way from a gradient, and it exists through its connections. In a world that can change in any direction, it can count on lasting only by changing with it, through feedback that arrives along its connections. Feedback plus retention is a learning loop, and when kept changes also raise the surplus, the loop becomes the vortex. What retained organization opens up next, the accessibility around which the earlier cores were built, is how the vortex turns. In a network, the whole's balance is the commons, the vortex of the whole: a node can live off its interest, its own creation plus what the commons regenerates, but taking more eats the principal, and not cheating gains more over time. The title's simulation of reality means the internal model through which a system experiences and acts in the world; it is not a claim that reality itself is simulated. The theory is a candidate architecture, not an established empirical law; CORE.md says what is proved, what is assumed and what is open. Its intelligent-system specialization adds one premise-free structural anchor — claims that exclude each other cannot all be true — and derives machine-checked constraints on how a learning system can change without making its remaining errors structurally undiscoverable. This is independent work by Albert Jan van Hoek; interpretations and applications are the author's own and are not institutional positions.
The world model. Every intelligence acts through world models; having a functional one takes work. A functional world model is held, for a stated scope and use, with the aim of fitting how things are, tested against evidence it cannot fix in advance, as sharp as that evidence supports subject to calibration, and correctable. In a collective it depends on two things: what contributors pass on, and how their contributions are combined. This is what the title asks, and since v23.0 it opens CORE.md. The full account, revised after three independent reviews, is World models, and what makes one functional; its narrow formal core is WorldModel.lean, 30 checked results, of which five are in the core.
Orientation. When answers become cheap, the bottleneck moves from finding solutions to taking them up. Yesterday's aims become today's tools, and prediction becomes a tool to change what it predicts. This repository is written in that spirit: not to predict the future, but to help build it. That is a stance, not a result; see Building the future.
| If you want to… | Go to |
|---|---|
| see the core dynamic on one page | CORE.md, checked in one Lean file: EbECore.lean |
| see what the framework predicts, and how it could be wrong | PREDICTIONS.md |
| explore the health of intelligence, and of the commons it lives in (concept notes) | research/health-of-intelligence/ |
| see how the author and an AI assistant build this model together, where they disagreed and what decided it | DIALOGUE.md (AI assistants: start with CLAUDE.md) |
| understand the central argument quickly | SCAP Seed, then Anchor-Safety |
| go deeper into the universal theory | THEORY_CORE_V21.md and FORMAL_THEORY_MAP.md: the v21 statement, centred on accessibility, which CORE.md reframes around persistence |
| read the book sources | Chapters/ and the online book edition |
| check the proofs | Verify below |
| see what is proved, assumed and open | CLAIMS_V21.md, ANCHOR_SAFETY.md, SELF_MODEL.md |
| start from the smallest replayable formal object | scap-seed/ |
| browse the papers | papers/ |
| listen or watch | Listen and watch below |
| cite the project | Cite below |
The full statement, with every step's proof, is CORE.md. It opens with what the theory is about:
- World models, and what makes one functional: held to fit how things are, tested, as sharp as the evidence supports, and correctable; in a collective, contributors and the way they are combined both matter.
- Existence is paid for from a gradient: slack is uptake minus upkeep.
- Existence through connections: delete the return edge and the node it fed declines; a defector that cannot live alone gains at first and then falls with its partner (proved for two nodes without a new partner; with partner switching, until detection and reputation make the next partner cost more than defection saves); the same structure appears at every level, inward and outward.
- Selection comes free: what pays its way lasts; selection selects persistence, not goodness.
- A changing world requires reconfiguration: a sealed configuration cannot count on lasting; one that follows reliable feedback stays in step.
- The learning loop becomes the vortex: kept changes that pay fund more trying and widen what can be tried.
- The commons is the vortex of the whole: live off the interest, not the principal; capture pays at first, not cheating gains more over time, and rules make it pay now.
- Boundaries: the vortex can stop.
- The same model across domains, with intelligence worked out formally, as below.
- The anchor. Mutually incompatible live claims cannot all be true; while a live rival remains, certainty from inside a model is not itself a certificate of correspondence with reality. See TheRoom.lean and the SCAP Seed.
- Common ground. Before agreement, informative content can itself be rivalled; the shareable structure is the maintained possibility of correction rather than a proposition all sides already accept. See Alignment.lean.
- Anchor-safety. Learn, act and commit, but do not make a still-live error in a retained commitment structurally undiscoverable. See ANCHOR_SAFETY.md.
- Self-model. Claims a system makes about itself receive no epistemic exemption from claims about the world. See SELF_MODEL.md.
- Shared layer. A shared representation can help as a connector of independent views; if it becomes their common determinant, its blind spots can propagate through the network. This is an information-theoretic result, not a claim about nationality, ownership, training provenance or benefit distribution. See GlobalLayer.lean and SharedLayerDynamics.lean.
- Transitions. A learning network may forget, compress, re-encode, rewire and change members provided blind live cuts do not expand. See CorrectionTransition.lean.
A healthy intelligence, individual or shared, does not need its parts to stay unchanged; it needs change never to erase the last way of discovering a still-live error.
What else a healthy intelligence needs (grounding, learning that keeps what still works, affordable upkeep, recovery) and what makes the commons around it healthy are developed as concept notes in research/health-of-intelligence/.
- The core on one page: CORE.md — the opening on world models and the eight steps from the ledger to the commons, with every result collected and checked in formalization/ebe-core/EbECore.lean.
- The full v21 statement of the universal theory (centred on accessibility; read its framing through CORE.md): THEORY_CORE_V21.md, FORMAL_THEORY_MAP.md, FORMAL_THEORY_ENDPOINT.md, and the specialization seam UNIVERSAL_TO_INTELLIGENCE.md.
- Universal Lean formalization: formalization/ and especially formalization/cumulative-accessibility/.
- Cumulative Reproduction Model: research/cumulative-reproduction/ — production, loss, resource dynamics and numerical experiments.
- Health of intelligence and of the commons: research/health-of-intelligence/ — concept notes, not results: a shared measure of the health of intelligence for people, AI systems and institutions, the health of the commons they live in, the research briefs and reports behind them, and their review.
- Building the future: research/building-the-future/ — an essay, not a result: what changes when answers become cheap, and why responsibility shifts toward uptake.
- Transparent pricing: research/transparent-pricing/ — an application to economics: access at cost, a separate bounty for solving problems, every choice visible; with a standalone Lean package in formalization/transparent-pricing/.
- World models, and what makes one functional: research/world-model/ — a definition note: world model and functional world model, sharpness subject to calibration, correctability, and what a collective world model needs from its contributors and from how it combines them; with independent reviews and the responses; and a companion on the other models an intelligence keeps in order to act and persist.
- Cooperative error identification: research/cooperative-correction/ — conjectures under investigation, revised after an independent review and an audit: whether different world models can expose each other's errors without a shared worldview, build a shared standard of evidence through interaction, and improve together, how much diversity of methods to keep, and when to stop investigating and act; with formal models and seeded simulations.
- Integrity of world models: research/world-model-integrity/ — a concept note: deliberate manipulation by words or by devices as two routes to one harm, and the neurorights read as part of existing human rights, which name principles and set boundaries.
- Anchored Correctability: research/anchored-correctability/ — semantics, tracking, networks, persistence, SCAP, alignment, Anchor-Safety, self-model, shared layers and correction-preserving transitions.
- Portable seed: scap-seed/ — a smaller self-contained Lean project with its own ledger and simulations.
- Independent verification material: verification/.
- Papers: papers/.
- Book: Chapters/, Backmatter/,
main.tex, and the website sources under docs/ with mkdocs.yml.
Nothing in this release deletes, moves, renames or rewrites the historical corpus. The older cores and ledgers carry a short pointer to CORE.md at the top. Nothing in the archive is a claim of the current theory unless CORE.md states it.
- THEORY_CORE_V17.md — earlier recursive-organization core; superseded as the repository front door by the v21 universal synthesis.
- THEORY_CORE_V20.md — frozen retained-organization/accessibility peer-review core; inherited by later releases.
- THEORY_CORE_V21.md — the full v21 statement of the universal theory, centred on accessibility and inherited unchanged by v22; CORE.md reframes it around persistence, and its results remain valid.
- THEORY.md, DYNAMIC_OVERVIEW.md, APPLICATION_MAPPINGS_V17.md — broader earlier syntheses and mappings; retained for lineage.
- Individual_essays/ and Original linkedIN posts/ — essays and original posts, kept as written. Some essays were written by AI systems in conversation with the author, as their titles say; they are part of the record, not evidence for the theory.
- ebe_science_religion_synthesis.md — an earlier reflection on science and religion; interpretation, not a claim of the theory.
- concepts.json — the v17 concept graph, not updated since.
- Discovarian_creed.tex, Discoverian_creed_better.tex, Discoverinan_creed_better_improved.tex — historical creeds, kept as written. The current creed, with each line marked by what it rests on, is in research/discovarianism/.
- Presentations/ — presentation sources, kept as written; among them Kwalitijd (October 2026, in Dutch), the story of this repository told through quality of life, looking ahead and choosing together.
- Rendered and source papers under papers/ — including historical PDFs; source lineage is preserved rather than normalized.
- RESEARCH_GUIDE.md provides a wider map of the corpus.
The current Lean toolchain for the principal formal packages is leanprover/lean4:v4.33.0.
The core in one file — the 96 results behind CORE.md, with every statement and axiom set printed (also usable as a Lean dependency; see formalization/ebe-core/README.md):
cd formalization/ebe-core
lake update
lake exe cache get
lake build EbECore
lake env lean EbECore.leanUniversal retained-organization/accessibility core:
cd formalization/cumulative-accessibility
lake update
lake exe cache get
lake build CumulativeAccessibility.AuditAll CumulativeAccessibility.VerificationSurface
lake env lean CumulativeAccessibility/VerificationSurface.leanDeep intelligent-system package:
cd research/anchored-correctability/lean
lake buildIts audit surfaces are in lean/AnchoredEvolution/*Audit.lean. CI rejects sorryAx and pins these counts:
| Audit | Headline results |
|---|---|
Audit.lean |
227 |
AnchorSafetyAudit.lean |
16 |
SelfModelAudit.lean |
11 |
GlobalLayerAudit.lean |
4 |
SharedLayerDynamicsAudit.lean |
11 |
CorrectionTransitionAudit.lean |
16 |
Within the 16-result Anchor-Safety audit, 15 results are axiom-free; anchorSafe_iff_cuts_separated uses classical logic in one direction. Other older formal surfaces have their own explicit axiom reports; machine checking proves implications under stated premises, not empirical truth.
Cumulative Reproduction Model:
cd research/cumulative-reproduction/lean
lean CumulativeReproduction.leanSCAP Seed:
cd scap-seed
scripts/verify.shThe seed audits 184 headline results. The integrated release matrix is .github/workflows/v21-integration-check.yml; the deep package has its own anchored-correctability workflow.
| Status | What it means here |
|---|---|
| Proved | Lean-checked implications under the definitions and hypotheses in the formal files; axiom status is printed in the audit files. |
| Assumed | Named premises such as liveness, open change, misfit costing slack and fit paying its way (from which correctability is derived; older files state the correctability aim directly), selection falling with cost, a node's return share and a commons that regenerates with its stock, evidence reliability where invoked, resource bounds, and model-to-domain mappings. |
| Interpretation / open | Empirical universality, whether real humans/organizations/models instantiate the predicates, numerical identity, causal social interpretations, political/normative conclusions, and other domain claims not established by the formal proofs. |
Full ledgers: CORE.md for the core, CLAIMS_V21.md, ANCHOR_SAFETY.md, SELF_MODEL.md, and scap-seed/CLAIMS.md.
Open problems remain visible rather than being converted into claims: the reachability–independence trade-off (including the Zollman-style concern that more connectivity can erase independent exploration), decoding and trust, total rather than merely pivotal contribution, the full upkeep/responsibility argument, and empirical tests of whether real updates are correction-preserving.
The work also exists as music, spoken word and video:
- Emergence on SoundCloud: https://soundcloud.com/emergence-223803727
- Autonomous Interdependence on YouTube: https://www.youtube.com/@AutonomousInterdependence
- Online book edition: https://albertjanvanhoek.github.io/Evolution-by-Emergence/
The website is the historical online book edition; this README is the current repository front door.
The long-lived Zenodo DOI used across the release lineage is 10.5281/zenodo.15207807. Cite the exact tagged release/Zenodo version when reproducibility requires a fixed snapshot; Zenodo assigns the v23.0 version DOI after archival.
@software{vanhoek_welcome_real_world_v22_3,
author = {van Hoek, Albert Jan},
title = {Welcome to the Real World: How You Live in Your Own Simulation of Reality — and How We Can Still Share a World. A Theory of Persistence, Emergence, Learning, and Correctable Intelligence},
version = {v23.0},
year = {2026},
doi = {10.5281/zenodo.15207807},
url = {https://github.com/albertjanvanhoek/Evolution-by-Emergence}
}Plain text:
van Hoek, Albert Jan. (2026). Welcome to the Real World: How You Live in Your Own Simulation of Reality — and How We Can Still Share a World. A Theory of Persistence, Emergence, Learning, and Correctable Intelligence (v23.0). Zenodo. https://doi.org/10.5281/zenodo.15207807
See CITATION.cff for machine-readable citation metadata. The project was developed with substantial AI-assisted drafting, critique and formalization support; formal claims are checked by Lean, while authorship and responsibility for the released work remain with the named author.
Except where a file or third-party notice says otherwise, original repository material is available under CC BY 4.0 OR Apache-2.0; choose either license. See DUAL-LICENSING.md, LICENSE-APACHE-2.0, and License.
Some subprojects or third-party materials carry their own notices; those local terms remain authoritative.