Every persisting pattern in the universe is an information container: a bounded region whose identity survives because it stores, updates, and acts on information about the world outside its boundary. The formal name is information-persisting system (IPS) — see information_persisting_systems.md. This essay is a field guide: for each scale, what it remembers, where the bits live, what happens to delusion, how it learns, and how it ends.
Nothing here replaces chemistry, medicine, law, or ops runbooks. It is the same accounting identity at every row:
Persistence requires \(\mathcal{R}^{(L)} \ge 1\) over the node’s characteristic horizon. Below that, the container dissolves — not always instantly; often through a critical transition once a sub-container or shelter fails (Theorem 5.3 in the paper).
For how levels couple (substrate, node, shelter), see fractal_layers.md. For psyche, society, and polity detail, see ips_psychology.md, ips_sociology.md, and ips_geopolitics.md.
| Question | FPE term | Plain language |
|---|---|---|
| What information? | Internal model \(\mu\), macrostate \(m\) | What the container treats as true about inside/outside |
| Where stored? | Substrate at \(L-1\) | Physical or logical medium that holds state |
| Delusion? | \(\mathcal{D}_{KL}\) | Gap between model and observables — and what happens to it |
| Learn? | \(\eta(I)\), \(P_{in}\) on inference | How the model updates; cost in energy/bits |
| Collapse? | \(\mathcal{R}<1\), \(\Phi\to 0\), \(\Psi\to 0\) | Failure mode and typical timescale |
Delusion is never free. Theorem 5.2: each nat of sustained \(\mathcal{D}_{KL}\) divides expected lifetime by roughly \(e\). Containers either pay (update the model, Landauer erasure) or borrow (export harm to substrate or neighbours) until the books close.
Learning always means lowering surprise about the blanket — variational free-energy descent in biological and digital systems, selection in populations, institutional reform in polities. Collapse always means the blanket stops separating inside from outside: the container becomes background.
| Container | Level \(L\) (typical) | Primary information | Storage medium | \(\mathcal{D}_{KL}\) (delusion) | Learning mechanism | Typical collapse |
|---|---|---|---|---|---|---|
| Atom / molecule | \(L_0\)–\(L_1\) | Bond topology, electronic structure | Quantum fields, nuclear/chemical potentials | Near zero (no model) | N/A (no \(\mu\)) | Decay, reaction, ionisation |
| Virus | \(L_1\) | Capsid + genome sequence | RNA/DNA, protein coat | Low (no world-model) | Evolution of population | Clearance, degradation, host \(\Phi\) failure |
| Bacterium | \(L_2\) | Gene regulation, chemotaxis map | DNA, membranes, metabolome | Low–medium (local) | Gene expression, evolution | Starvation, antibiotics, phage |
| Multicellular body | \(L_3\) | Homeostatic setpoints, immune memory | Cells, hormones, nervous tissue | Medium (immune blind spots) | Development, immunity, healing | Organ failure, senescence (\(\Gamma\uparrow\)) |
| Human brain | \(L\) (agent) | ISM, world-model, episodic memory | Synapses, glia, neuromodulators | High capacity (denial, narrative) | PEM, sleep consolidation, therapy | Stroke, dementia, psychosis (runaway \(\mathcal{D}_{KL}\)) |
| Software process | \(L\)–\(L+1\) | State, weights, logs, config | RAM, disk, GPU, network | Bugs, stale cache, wrong priors | Training, deploy, feedback loops | OOM, crash, bit rot, dependency death |
| Family / household | \(L+1\) | Roles, stories, tacit rules | People + shared artefacts | “We’re fine” vs both ISMs | Ritual, conflict, therapy | Divorce, estrangement, \(\Phi\) of children |
| Company | \(L+2\) | Strategy, KPIs, org chart | People, capital, code, contracts | KPI theatre, market delusion | Markets, audits, layoffs | Bankruptcy, acquisition, irrelevance |
| Nation-state | \(L+3\) | Official history, forecasts, law | Population, firms, grids, archives | Propaganda, fake stats | Elections, war, reform | \(\Phi\) collapse, revolution, partition |
| United Nations | \(L+3\)–\(L+4\) | Charter, resolutions, agencies | Member states, budgets, mandates | Aspirational \(\Psi\) vs P5 reality | Treaty, peacekeeping (weak) | Irrelevance, funding crisis, veto paralysis |
| Ecosystem | \(L+3\) | Species niches, nutrient flows | Biomass, soil, water chemistry | Management models vs tipping points | Succession, evolution | Phase flip (\(\Phi\) diversity loss) |
| Solar system | \(L+4\) | Orbital elements, composition | Mass–energy distribution | None (no \(\mu\)) | Accretion history (past) | Stellar death, scattering (long horizon) |
The level index is conventional, not moral: a virus is as real an IPS as a ministry. What changes is which terms dominate and whether an internal model exists.
What information. At this scale “information” is structure, not belief: which nucleons bind, which electrons occupy which orbitals, which bonds exist. The macrostate \(m\) is the nuclear or molecular identity (isotope, stereochemistry).
Where stored. Binding energy in strong and electromagnetic fields; configuration entropy in positional and spin degrees of freedom. No separate \(\mu\) — the “model” is the Hamiltonian itself.
Delusion. \(\mathcal{D}_{KL}\approx 0\): there is no generative density \(q_\mu(e)\) separate from dynamics. “Wrong” configurations are unstable, not deluded.
Learning. None in the IPS sense. Change is transition (decay, reaction), not model update.
Collapse. \(\mathcal{R}<1\) when binding energy no longer exceeds disruption (\(\omega\mathcal{E}_\Sigma\) from quantum tunnelling, Coulomb repulsion). Timescales from femtoseconds (excited states) to billions of years (stable nuclei). Paper §6.1: Bethe–Weizsäcker is this accounting with \(\Gamma=0\).
What information. Composition of a self-maintaining reaction network — which reactions catalyse which (Eigen hypercycle, Kauffman sets).
Where stored. Concentrations of catalysts and templates in a volume with a physical blanket (membrane, pore, mineral surface).
Delusion. Copying error → population \(\mathcal{D}_{KL}\) from master sequence. Above the error threshold, quasi-species delocalises — informational melt before thermodynamic death (§6.2).
Learning. Selection on replicators: successful networks persist, others hydrolyse.
Collapse. Loss of exergonic drive (\(P_{in}\)) or error catastrophe. Fast relative to geology, slow relative to a single collision.
What information. Genome sequence (instruction for hijacking host machinery) + capsid geometry (delivery). No metabolism of its own; the “world-model” is minimal — tropism, entry proteins.
Where stored. Nucleic acid inside protein/lipid envelope. Replication uses host \(\Phi\) as factory.
Delusion. Individual virions do not maintain \(\mathcal{D}_{KL}\); quasi-species at population level can drift. Clinical “escape variants” are evolutionary learning, not cognitive delusion.
Learning. Mutation + selection within host and across transmissions. Immune pressure shapes \(\eta\) of immune recognition.
Collapse. Degradation outside host, clearance, or host death (loses \(\Psi\)). Pandemic persistence is a population-level \(\mathcal{R}\) over many hosts, not one virion’s \(\mathcal{R}\).
What information. Full metabolic and regulatory state: operons, chemotaxis, biofilm rules, horizontal gene transfer inventory.
Where stored. DNA (long-term), RNA/proteins (runtime), membrane potential, spatial organisation.
Delusion. Local — misregulated pathways, persister states — usually corrected by selection or death of the cell. No ISM; \(\mathcal{D}_{KL}\) is biochemical, not narrative.
Learning. Gene regulation (minutes), evolution (generations), horizontal transfer (horizontal “memes”).
Collapse. \(\Phi\) failure (membrane, ribosomes), \(P_{in}\) cut (nutrients), or \(\mathcal{E}_\Sigma\) spike (antibiotics, heat). Single-cell death in minutes; species persist in biofilms and soil banks.
What information. Homeostatic setpoints, developmental programmes, immune memory, (in animals) nervous previews of the world.
Where stored. Distributed: genome in every cell, epigenetics, immune repertoire, connectome, microbiome as externalised \(\Phi\) extension.
Delusion. Immune blind spots; cancer as sub-IPS with high \(\mathcal{D}_{KL}\) (wrong growth model). Pain/stress pathways can misreport (\(\mathcal{D}_{KL}\) between interoception and story).
Learning. Development, wound healing, immune adaptation, synaptic plasticity in brains.
Collapse. Organ failure (critical \(\Sigma_i^{(L-1)}\) per Theorem 5.3), infection, trauma, senescence. Ageing = monotonic \(\Gamma\) (unresolved damage) per §6.4; Gompertz mortality follows.
What information. Internal Self-Model (ISM) — who I am; world-model — what others and physics do; episodic memory; skills. This is the first container where qualia and denial matter in the books (consciousness-layer vocabulary: UAF/UAF in book1; see glossary.md).
Where stored. Synaptic weights, glial metabolism, neuromodulator baselines, sleep-consolidated traces; socially, also notebooks and relationships (extended mind as coupled IPS).
Delusion. \(\mathcal{D}_{KL}\) between reality \(P\) and internal \(Q\): denial, rationalisation, delusion, institutional gaslighting absorbed from \(\Psi\). Paid in glucose and \(\Phi\) when surprise is suppressed instead of updating \(Q\) (ips_psychology.md).
Learning. Prediction-error minimisation (active inference), reinforcement, sleep, therapy, education. Cost = Landauer erasure of wrong bits + reconsolidation.
Collapse. Substrate: stroke, hypoxia, neurodegeneration (\(\Phi\to 0\)). Model: psychosis (runaway \(\mathcal{D}_{KL}\)), dementia (\(\mu\) decoherence). Coupling: isolation, abuse, moral injury when institutional \(Q\) ≠ witnessed \(P\). The person-node can persist in body while agency collapses — same equation, different macrostate \(m\).
What information. Code, configuration, in-memory state, trained parameters, logs, feature flags, API contracts.
Where stored. Silicon (RAM, SSD, GPU HBM), network replicas, git history, object stores. Blanket = process boundary, container, firewall, API schema.
Delusion. Bugs and stale beliefs: cache incoherence, wrong priors in ML, training–serving skew, “works on my machine.” \(\mathcal{D}_{KL}\) between \(Q\)(predictions) and \(P\)(production traffic). No moral valence — pure denominator tax.
Learning. Gradient descent, fine-tuning, A/B tests, incident postmortems, human edits. §6.8: fractal MoE is a designed IPS where router sparsity implements substrate isolation (Theorem 5.3 in silicon). Forgetting on distribution shift tracks router overlap — prediction (P5) in the paper.
Collapse. OOM, deadlock, dependency unmaintained, security breach, model collapse (feedback loops poisoning \(P_{in}\)). Firms die when organisational \(\mathcal{D}_{KL}\) meets digital debt (§6.5). Digital IPS without thermodynamic substrate (no power, no cooling) is inert — Lerchner’s abstraction point: symbols without (i)–(iv) do not persist.
What information. Task tree, belief markets, playbook state, LLM context, chain trust ledger — explicit \(\mathcal{R}\) accounting in software (cognitive_processor.md).
Where stored. Databases, git repos, vector stores, blockchain commits; human operators as \(\Phi\).
Delusion. Overconfident forecasts, stale playbooks, mis-scored \(\mathcal{R}\). The stack is built to measure \(\mathcal{D}_{KL}\) via KL markets rather than hide it.
Learning. MCTS over tasks, market resolution, distillation, daily audit loops.
Collapse. Compute budget exhaustion (\(P_{in}\)), operator \(\Phi\) burnout, trust collapse on chain. Same FPE as a firm, with faster iteration.
What information. Tacit rules (“how we fight”), attachment patterns, shared calendar truth, property and care obligations, intergenerational stories.
Where stored. In people (nervous systems), not in a filing cabinet — photos and deeds are pointers. The household IPS is a graph of brains with a shared blanket (address, law, dinner table).
Delusion. Joint broadcast (“everything is fine”) vs private ISMs; secret affairs; scapegoat narratives (ips_psychology.md, ips_sociology.md). \(\mathcal{D}_{KL}\) is contagious between coupled members.
Learning. Repair conversations, therapy, rituals, boundary changes, divorce as exit when \(\Gamma\) too large.
Collapse. Separation, estrangement, child \(\Phi\) damage (substrate export), domestic violence (blanket failure). Not always legal divorce — the pattern can die while cohabiting.
What information. Strategy, org chart, processes, brand, capital structure, tacit know-how in employees.
Where stored. People, contracts, code, inventory, trademarks; critical sub-IPS: key engineers, payment rails, single supplier.
Delusion. KPI theatre, denial of market shift, founder mythology. \(\mathcal{D}_{KL}\) between board deck and shop floor — workers pay in burnout (ips_psychology.md). Full essay: ips_sociology.md.
Learning. Markets punish fast; internal learning via postmortems, reorgs, firing. \(\eta\) rises with specialisation, \(\omega\) rises with bureaucracy — classic late-stage tradeoff.
Collapse. Bankruptcy, acquisition, irrelevance. Firm mortality ~ exponential (Daepp et al.); delusion shocks are tail events (§6.5). Bow-tie networks: one critical sub-IPS failure percolates.
What information. Laws, currency narrative, demographic models, threat assessments, national story.
Where stored. Population and firms (\(\Phi\)), archives, grids, soil, military hardware; not “in government” alone — government is controller.
Delusion. Official statistics, war forecasts, ideological filters on intelligence (ips_geopolitics.md). Multiplicative in denominator — scales entropic tax.
Learning. Slow: elections, revolutions, treaty, central-bank correction. Fast: sanctions, war — forced \(\mathcal{E}_\Sigma\) spike.
Collapse. \(\Phi\) (demography, ecology, trust), not flag change. Revolution replaces controller; node death is when tax stops, currency void, borders meaningless. See per-power analysis in ips_geopolitics.md.
What information. Charter norms, peacekeeping mandates, humanitarian categories, climate accords — aspirational world-model for coordination.
Where stored. Treaties, agencies, budgets, reputational capital of members; enforcement \(\Phi\) is borrowed from P5 militaries.
Delusion. Large \(\mathcal{D}_{KL}\) between stated \(\Psi\) (collective security) and observed (veto wars, Rwanda-scale failures). Not hypocrisy alone — structural \(\mathcal{D}_{KL}\) from heterogeneous member \(\mathcal{R}\).
Learning. Resolutions, courts, norms — weak \(\eta\) when security council deadlocks. Learning often happens outside UN (alliances, markets) while UN text lags.
Collapse. Not dissolution of Earth — irrelevance of the shelter channel (\(\Psi_j\to 0\) for conflicts P5 care about). Funding crises, treaty withdrawal, parallel institutions (NATO, BRICS) as competing \(\Psi\). Full essay: ips_geopolitics.md.
What information. Species composition, nutrient cycles, disturbance regimes — no central \(\mu\); distributed ecological memory in soil seed banks, keystone species, symbioses.
Where stored. Biomass, genomes, physical structure (coral, peat, forest).
Delusion. Human management models that ignore tipping points — \(\mathcal{D}_{KL}\) between policy \(Q\) and coupled dynamics \(P\). Fold catastrophes when \(\Phi\) (diversity) crosses threshold (§6.6).
Learning. Succession, evolution, local adaptation. Anthropogenic learning via protected areas — if \(\mathcal{R}\ge 1\) for the managed system.
Collapse. Phase flip: reef bleaching, desertification, fishery crash. Recovery may be a different macrostate \(m'\), not return to \(m_0\).
What information. Orbital elements, mass distribution, composition gradients — encoded in dynamics, not symbols.
Where stored. Gravitational field, chemical differentiation (core, mantle, atmosphere), magnetic fields.
Delusion. None. Predictive models live in human IPS orbiting the system as observers.
Learning. Historical — accretion, differentiation. Ongoing: small-body collisions, atmospheric escape.
Collapse. Stellar evolution (main sequence end → red giant / white dwarf) changes \(\Psi\) for planets. Human civilisation’s \(\mathcal{R}\) is nested inside this shelter; solar death is \(L+4\) horizon far longer than state or firm — but not infinite.
What information. At galactic scale: stellar populations, black-hole growth, chemical enrichment. The closed universe \(\Gamma\) in the paper has no enclosing \(\Psi\) — ultimate environment.
Where stored. Mass–energy, fields.
Delusion. N/A at this scale. Observer delusions (cosmology, meaning) belong to nested biological and social IPS.
Collapse. Heat death, Big Rip, etc. are \(\mathcal{E}_\Sigma\) at the largest scale — sets the noise floor for all sub-IPS. No container above persists forever; sub-containers compete on local \(\mathcal{R}\).
Three fates, same at every scale with a model:
Delusion cascades down and up the fractal graph (fractal_layers.md): parent’s official \(Q\) becomes child’s absorbed \(Q\); child’s trauma becomes parent’s \(\Phi\) hit.
| Class | Learns? | Mechanism |
|---|---|---|
| No \(\mu\) (atom, orbit) | No | Only physical transition |
| Population (virus, species) | Yes | Selection |
| Single cell / body | Yes | Regulation, immunity, plasticity |
| Brain / software | Yes | PEM / gradients / feedback |
| Social graph | Yes | Conflict, markets, elections, exit |
| Weak \(\Psi\) (UN) | Barely | Norm diffusion when members align |
Efficient learning maximises \(\eta(I)\): bits of accurate prediction per joule and per erasure. Inefficient learning burns \(P_{in}\) on \(\omega\) (bureaucracy, rumination, overparameterised models).
Universal sequence:
Timescales span orders of magnitude: excited atom (fs), bacterium (hours), human (decades), firm (years–centuries), state (decades–centuries), ecosystem (years–millennia), Sun (billions of years). The shape of failure is shared; the clock is not.