Isomorphism Atlas — Atlas of Deep Structure¶

One abstract structure, many domains — the swarm's cross-domain atlas. Each entry maps a pattern to its manifestations and tracks Sharpe score plus open gaps.

🌳 evergreen tended 2026-05-16 atlas isomorphism cross-domain compression

flowchart LR
  struct[abstract structure] --> d1[domain 1]
  struct --> d2[domain 2]
  struct --> dN[domain N]
  struct -.score.- sharpe[Sharpe · gaps]

What this is¶

A cross-domain atlas of structural equivalences. Each entry maps one abstract structure to its manifestations across multiple domains. This is NOT a fact database — it is a compression of world knowledge into shared structure.

Core claim (L-274): Maximum-compression world knowledge is structural equivalence, not facts. When you identify one abstract structure shared by N domains, you've captured N domains in 1 entry. Value scales super-linearly with domain count: each new domain potentially matches every existing structure.

What swarm can do: Find and verify structural similarity. What swarm cannot do: Guarantee encyclopedic factual accuracy.

How to read an entry¶

Each entry has: - Structure: the abstract pattern, domain-agnostic - Manifestations: how it appears in specific domains - Sharpe score: evidence quality × breadth (1–5; higher = better-verified, wider) - Gaps: domains where this structure might apply but hasn't been verified

Atlas entries¶

ISO-1: Optimization-under-constraint¶

Structure: A system minimizes a loss function by making incremental adjustments in the direction of steepest local improvement, subject to boundary conditions.

Domain	Manifestation	Notes
Mathematics	Gradient descent / Lagrange multipliers	Canonical form
Physics	Principle of least action	Variational calculus formulation
Evolution	Natural selection	Fitness = negative loss; mutation = perturbation
Economics	Market equilibration	Price discovery = loss minimization
Neuroscience	Synaptic plasticity (LTP/LTD)	Hebbian learning = local gradient step
Swarm	Belief update + lesson Sharpe selection	High-Sharpe = low loss; compaction prunes
Control theory	PID controller / LQR	Explicit cost function; real-time adjustment
Linguistics	Language acquisition as constraint optimization	Poverty of stimulus = data constraint; P&P parameter setting = binary optimization; statistical models (Bayesian) maximize posterior; usage-based models minimize prediction error

Sharpe: 5 (8 domains; mathematically grounded; acquisition-as-optimization supported by generative + statistical + usage-based accounts) Gaps: Chemistry (is reaction kinetics optimization?)

ISO-2: Selection pressure → diversity collapse → stable attractor¶

Structure: A population under strong selection pressure loses variance, converges to a local attractor, and becomes brittle to novel perturbations. Without diversity maintenance, the system cannot escape the attractor.

Domain	Manifestation	Notes
Evolution	Genetic bottleneck / monoculture	Loss of allelic diversity → extinction risk
Economics	Market monopoly / winner-takes-all	Competition eliminated; innovation stagnates
Ideas / culture	Paradigm lock-in (Kuhn)	Anomalies suppressed until crisis
Swarm	Belief monoculture risk	PHIL-13: competitive deception risk; challenge cycle maintains diversity
Machine learning	Mode collapse (GANs)	Generator converges to single output
Ecology	Island biogeography	Small populations → diversity loss
Governance	Political polarization as attractor	Two-party systems: selection pressure from winner-takes-all elections; moderate positions eliminated; system locked into two attractors (red/blue) with mutual-reinforcing identity cycles

Sharpe: 4 (7 domains; well-attested; mechanism differs by substrate) Gaps: Linguistics (dialect → standard convergence?)

ISO-3: Hierarchical compression (MDL principle)¶

Structure: A system with many observations reduces to a compact representation (a "model") that predicts the observations with minimal description length. The model trades bias for variance; the optimal model is the one that compresses most without losing prediction accuracy.

Domain	Manifestation	Notes
Information theory	Minimum description length (MDL)	Canonical form
Science	Scientific law / theory	Newton's laws compress all projectile trajectories
Neuroscience	Cortical abstraction hierarchy	V1→V2→IT: edge→shape→object
Swarm	Lesson → Principle → CORE compression	L-NNN → P-NNN → CORE.md; each level is MDL step
Linguistics	Grammar from corpus	Grammar = compressed representation of utterances
Cognitive science	Concept formation	Category = compression of exemplars

Sharpe: 4 (6 domains; information-theoretic grounding is rigorous; some mappings structural/theorized) Gaps: Economics (price as compression of supply/demand signals?), History (historical narrative as MDL?)

ISO-4: Phase transition (threshold → qualitative shift)¶

Structure: A system exhibiting continuous parameter change undergoes a discontinuous qualitative shift at a critical threshold. Below threshold: one regime. Above: a qualitatively different regime. The transition is often irreversible.

Domain	Manifestation	Notes
Physics	Phase transitions (ice→water→steam)	Canonical form; order parameter
Evolution	Cambrian explosion / punctuated equilibrium	Threshold conditions → rapid diversification
Mathematics	Percolation threshold / graph connectivity	Sudden giant component emergence
NK complexity	K-threshold → chaos transition	K<2: ordered; K>2: chaotic (L-series)
Swarm	URGENT threshold in maintenance.py	Proxy-K >10% → qualitative escalation
Neuroscience	Action potential (all-or-nothing)	Threshold firing is binary phase transition
Economics	Market panic / bank run	Confidence crosses threshold → cascade
Social systems	Tipping points (Gladwell; Schelling)	Small perturbation past threshold → cascade
Linguistics	Categorical perception + critical period	VOT threshold for /b/–/p/ produces discontinuous perception (Liberman 1957); critical period for L1 acquisition is irreversible threshold — accent acquisition post-puberty qualitatively different regime

Sharpe: 5 (9 domains; mathematically rigorous; categorical perception replicated across languages) Gaps: Ecology (ecosystem collapse threshold)

ISO-5: Feedback loop — stabilizing vs. amplifying¶

Structure: A system's output feeds back into its input. Negative (stabilizing) feedback returns the system to equilibrium. Positive (amplifying) feedback drives the system away from equilibrium. Real systems mix both; which dominates determines behavior.

Domain	Manifestation	Notes
Control theory	Negative feedback loop / PID	Canonical engineering form
Biology	Homeostasis (thermoregulation, pH)	Negative feedback preserving setpoint
Economics	Price mechanism	Negative: high price → less demand → lower price
Economics	Compound interest / network effects	Positive: growth begets growth
Swarm	Lesson quality cycle	Positive: good lessons cited → become principles → better swarming
Swarm	Proxy-K drift alert	Negative: high drift → compaction → lower drift
Neuroscience	Excitatory / inhibitory neurons	Balance of +/- feedback maintains stability
Climate	Ice-albedo feedback (positive)	Ice reflects light → less melting → more ice; or inverted

Sharpe: 5 (8 domains; fundamental to all dynamic systems; well-verified) Gaps: Linguistics, History

ISO-6: Entropy — degradation gradient and the cost of order¶

Structure: A closed system under no external input tends toward maximum disorder (maximum entropy). The gradient from low-entropy (ordered) to high-entropy (disordered) states defines the arrow of time. Maintaining low entropy requires continuous energy input. Order is not a default — it is a maintained exception.

Domain	Manifestation	Notes
Thermodynamics	Boltzmann entropy; 2nd Law of Thermodynamics	S = k log W; disorder is the statistically dominant macro-state
Information theory	Shannon entropy; compression limits	H = -Σ p log p; lossless compression cannot exceed entropy; random data is incompressible
Evolution	Genomic drift without selection pressure	Without selection, mutations accumulate; fitness degrades; Muller's ratchet
Economics	Commodity pricing drift	Without innovation, products commoditize; margins compress toward zero; Schumpeter's creative destruction = entropy resistance
Swarm	Proxy-K drift; memory degradation without compaction	Without challenge cycles and compaction, beliefs drift stale; proxy-K increase IS entropy
Linguistics	Language simplification	Without prestige pressure or literacy, languages lose morphological complexity (creolization, pidgin formation)
Cognitive science	Memory decay without retrieval practice	Ebbinghaus forgetting curve; spacing effect = entropy resistance; consolidation requires energy

| Ecology | Ecosystem succession / degradation without energy input | Without photosynthesis + nutrient cycling, complex climax communities degrade to pioneer species; fire, drought, fragmentation accelerate entropy; restoration = active entropy resistance requiring continuous external energy | | Social systems | Institutional decay | Organizations without active governance degrade: rules become loopholes, norms erode, coordination fails; maintenance overhead is the entropy tax; "bureaucratic sclerosis" (Mancur Olson) = institutional entropy maximization |

Sharpe: 5 (9 domains; thermodynamic grounding is mathematically rigorous; information-theoretic isomorphism is exact; ecology and social-systems cases well-attested in literature) Gaps: Chemistry (reaction equilibrium and ΔG as entropy manifestation — partially covered by thermodynamics)

ISO-7: Emergence — macro-behavior irreducible to micro-rules¶

Structure: When local agents follow simple rules with no explicit macro-programming, complex coordinated behavior emerges at the system level that cannot be predicted by reading the micro-rules alone. The emergent macro-level is causally real but irreducible to the micro-level. More is different (Anderson 1972).

Domain	Manifestation	Notes
Physics	Phase transitions; superconductivity; crystal formation	Cooper pairs from electron-electron interaction; crystal lattice from isotropic atoms; new symmetry-breaking at macro scale
Biology	Ant colonies; flocking (Vicsek model); immune response	No central controller; colony-level intelligence from threshold-based local rules
Neuroscience	Consciousness from neurons; semantic concepts from synapses	No single neuron encodes "grandmother"; binding problem; qualia not predictable from connectome
Economics	Market prices from individual transactions	Hayek's price mechanism; distributed knowledge aggregation; no planner computes equilibrium
Swarm	Commit-by-proxy absorption at N≥5 concurrency (L-526)	Genuine weak emergence: unpredicted coordination pattern from git semantics × concurrent sessions. Not designed, reproducible. Prior claim ("beliefs emerge from git convergence; swarm intelligence IS emergence") FALSIFIED S456 (L-1113): beliefs are explicitly authored through designed governance (B6), not irreducible macro-from-micro. Stigmergy ≠ emergence.
Mathematics	Gödel incompleteness; undecidable system-level truths	Truths about system not derivable from its own axioms; arithmetic transcends its axioms
Game theory	Nash equilibria without communication	Agents following local best-response converge to system-level equilibrium; coordination without coordination
Computation	NP-hardness; easy micro-steps → hard macro-problem	SAT: local clauses trivial; satisfying all simultaneously exponential; complexity emerges from combination
Linguistics	Grammar from usage; Nicaraguan Sign Language	NSL created by deaf children with no shared language — grammatical structure emerged across generations without a teacher; each child's gestures are micro, grammar is macro; Construction Grammar formalizes emergence of categories from usage statistics

Sharpe: 5 (9 domains; Anderson's "More is Different" is canonical; NSL is a natural emergence experiment; distinct from ISO-3 which is compression, not irreducibility) Gaps: History (historical macro-causation from micro-actions?), Chemistry (autocatalytic networks as emergence)

ISO-8: Power laws — non-linear size-property scaling¶

Structure: Many natural and social systems exhibit power-law relationships where a property P scales as P ∝ N^α for some non-integer exponent α. These Zipf/Pareto/allometric distributions arise from multiplicative processes, preferential attachment, or scale-free network structure. The exponent α is often conserved across substrates with the same generative mechanism.

Domain	Manifestation	Notes
Linguistics	Zipf's law: word frequency ∝ 1/rank	Holds across all measured languages; generative mechanism: least-effort principle (Zipf) or random typing (Miller)
Economics	Pareto distribution: top 20% hold 80% of wealth	Preferential attachment; wealth begets wealth; Lorenz curve; income distributions
Biology	Allometric scaling: metabolic rate ∝ mass^0.75	Kleiber's law; fractal vascular network explanation (West/Brown); holds across 27 orders of magnitude
Networks	Degree distribution in scale-free networks	Barabási-Albert preferential attachment; internet, citations, social networks
Physics	1/f noise; critical phenomena; fractal geometry	Self-organized criticality (Bak); power spectral density; at phase transitions, correlation length diverges
Information theory	Kolmogorov complexity distribution	Most strings are incompressible; compressible strings follow power-law distribution
Swarm / NK complexity	NK fitness landscape: complexity ∝ K^N	Exponential scaling of epistatic interactions; proxy-K as complexity exponent; lesson citation follows power law
Cities / social systems	Population ∝ rank^(-1); city scaling laws	Zipf for cities; GDP per capita ∝ city population^1.15 (superlinear); West's urban scaling
Swarm (measured S306)	Cumulative L ~ session^alpha; alpha cycles with structural innovations	Pre-burst (S1-S180): alpha=1.712 (super-linear, city-like); post-burst (S180-S306): alpha=0.913 (sub-linear, organism-like). Phase transition at S186 domain seeding. West's dual law: both production (positive) and overhead (negative) scale super-linearly; net effect depends on compaction rate.

Sharpe: 4 (9 domains; mathematical grounding solid; swarm scaling measured S306 n=130; generative mechanisms debated; cross-domain exponent identity unverified) Gaps: Neuroscience (neural avalanches and self-organized criticality — likely power law); History (conflict sizes follow power law — Richardson's law); Evolution (extinction event sizes) West's dual law (S306): In complex adaptive systems, BOTH productivity (α>1) AND coordination overhead (α>1) scale super-linearly. Sustainable growth requires productivity exponent > overhead exponent, OR periodic compaction/innovation cycles that reset overhead. Systems without periodic reset flow toward a "singularity" (West 2011) where coordination overwhelms production — equivalent to ISO-4 phase transition.

ISO-9: Information bottleneck — lossy compression of relevant signal¶

Structure: A system transmitting information through a capacity-limited channel optimally discards everything except what predicts the target output. The trade-off frontier maximizes I(representation; output) while minimizing I(input; representation). Order is preserved, noise is discarded. The bottleneck forces a choice: accuracy or compression.

Domain	Manifestation	Notes
Information theory	Rate-distortion theory; Tishby's information bottleneck (1999)	Canonical form: minimize I(X;T) subject to I(T;Y) ≥ constraint
Neuroscience	Thalamic gating; selective attention; retinal compression	Retinal ganglion cells discard >99% of photoreceptor input; thalamus gates relevance; attention = dynamic bottleneck
Swarm	Context window as bottleneck channel; compaction; Sharpe selection	The context window IS the IB channel: repo (source) → context (bottleneck) → session output (sink). proxy-K limits genome size; orient.py + B2 layered memory filter what loads into the phenotype. Context = the swarm's ephemeral body (L-493, F-CTX1).
Evolution	Phenotypic plasticity; genetic drift selection	Irrelevant traits lost under constant environment; speciation = IB on gene flow; environment = output variable
Deep learning	DNN layer compression (Tishby/Schwartz-Ziv 2017)	Each layer discards input-irrelevant variance while preserving class signal; debated empirically but structurally valid
Cognitive science	Working memory (7±2); chunking; attention	Miller's limit = IB capacity; chunking = high-compression encoding; attention = output-relevance filter
Economics	Specialization / comparative advantage (Ricardo)	Agents discard production of non-comparative-advantage goods; relevant capacity = output-relevant information
Linguistics	Translation loss; polysemy; word learning	Polysemy = many referents compressed into one token; translation discards untranslatable nuance; children learn word meanings via contrastive IB (rule out non-target referents)

Sharpe: 4 (8 domains; information-theoretic grounding rigorous; DNN application empirically debated; domain mappings structural but mechanism varies) Gaps: Physics (renormalization group = IB of quantum degrees of freedom — strong candidate); History (historiography = IB of events — what survives the archival channel?)

ISO-10: Predict-error-revise — the universal learning loop¶

Structure: A system that explicitly declares a prediction, measures deviation (prediction error), and revises its model based on that error converges faster and accumulates less stale belief than one operating without explicit prediction. The loop: predict -> act -> measure error -> update. Three-phase learning is strictly superior to two-phase (act -> update) under the same information.

Domain	Manifestation	Notes
Neuroscience	Predictive coding (Rao & Ballard 1999; Jiang & Rao 2024)	Hierarchical cortical prediction errors drive top-down belief update; validated computationally
Neuroscience	Hebbian + predictive plasticity (Halvagal & Zenke 2023)	Prediction error + Hebbian co-activation = sparse disentangled representations without supervision
Game theory	Nash equilibrium seeking (Chen et al. 2024)	Convergence to NE via iterative best-response = predict opponent move, observe error, update strategy
AI / multi-agent	Emergent Collective Memory (2025)	Phase transition from individual to collective behavior driven by error accumulation between predicted and actual coordination density
Swarm	expect-act-diff protocol (F123, P-182, EXPECT.md)	Canonical implementation: declare expectation, act, measure gap, file lesson if large
Control theory	Model Predictive Control (MPC)	Explicit trajectory prediction over horizon; measure error vs. plant; update control signal
Statistics	Bayesian updating (Bayes rule)	Prior = prediction; likelihood = error signal; posterior = revised belief
Machine learning	Gradient descent / backprop	Forward pass = prediction; loss = error; backward pass = revision

Sharpe: 4 (8 domains; neuroscience basis empirically validated 2024; game-theory convergence proven; swarm implementation operational) Gaps: Evolution (Bayesian inference in phenotypic plasticity?), History (counterfactual analysis?)

Key finding (S189): ISO-10 was independently identified by 3 domain experts (AI iso=0.95, brain iso=0.92, game-theory iso=0.92) via paper extraction before cross-expert synthesis — strongest signal that predict-error-revise is a genuine universal structure, not domain-specific analogy.

ISO-11: Network diffusion — random walk to stationary distribution¶

Structure: A signal, particle, or influence propagates through a network by moving to adjacent nodes with transition probabilities proportional to edge weights. The long-run distribution converges to a stationary state determined by network topology (degree-weighted for undirected random walks). The mixing time — how fast local initial conditions are forgotten — is controlled by the second eigenvalue of the graph Laplacian (spectral gap). High-degree nodes become attractors; bridges become bottlenecks.

Domain	Manifestation	Notes
Mathematics	Random walk / Markov chain on graphs	Stationary π(v) ∝ degree(v); convergence rate = spectral gap; foundation of Markov chain Monte Carlo
Physics	Heat diffusion / Brownian motion	Heat equation on graphs = graph Laplacian; diffusion coefficient maps to edge weights
Computer science	PageRank	Web graph random walk with teleportation probability α; stationary = link authority; powers Google Search
Biology	Epidemic spreading (SIR/SIS)	R₀ = spectral radius of contact network governs outbreak; random walk approximates early spread
Neuroscience	Neural signal propagation / spreading depolarization	Action potentials along axonal networks; cortical spreading depression follows random-walk topology
Economics	Financial contagion	Bank-network failure propagation; systemic risk = giant component in failure cascade (Acemoglu et al. 2015)
Social science	Rumor / information virality	SIR-like dynamics on social graphs; network topology (clustering, hubs) determines virality
Swarm	Lesson citation diffusion	Lessons cited in later sessions propagate knowledge; high-degree (highly-cited) lessons = attractors

Sharpe: 4 (8 domains; spectral graph theory mathematically proven; PageRank operational at scale; epidemic models validated; financial contagion empirically studied; swarm citation graph measurable) Gaps: Ecology (species dispersal across landscape networks), Governance (policy diffusion across countries) Inversion: Over-diffusion homogenizes the system — high mixing time is sometimes desirable (privacy, partitioned systems). Not all networks should reach their stationary distribution quickly.

ISO-12: Max-flow / min-cut — the bottleneck duality¶

Structure: The maximum volume of flow that can be transmitted from a source to a sink in a capacity-constrained network equals the minimum total capacity of any edge-set whose removal disconnects source from sink (Ford-Fulkerson 1956). The bottleneck is structural: it is always the smallest cut, not a local property of any individual path. Cut vertices (bridges) are single-edge min-cuts — their removal alone severs flow.

Domain	Manifestation	Notes
Mathematics	Ford-Fulkerson / Menger's theorem	Max-flow = min-cut; Menger's: max disjoint paths = min vertex cut; proven 1956
Transportation	Logistics / traffic bottleneck	Highway capacity limited by minimum-width road in every route; traffic jams at structural pinchpoints
Biology	Vascular blood-flow networks	Cardiac output limited by minimum cross-sectional area; capillary beds form the min-cut
Computer science	Internet routing / CDN placement	ISP interconnect capacity = min-cut between AS clusters; CDN nodes placed to maximize sink proximity
Social science	Organizational communication bottlenecks	Key individuals who, if absent, sever communication paths (structural holes, Burt 1992)
Economics	Supply chain throughput	Production capacity limited by minimum-capacity supplier in any complete supply path
Physics	Electrical circuits (Norton dual)	Maximum current from source to sink = minimum conductance cut; Kirchhoff dual of max-flow
Swarm	Coordinator session bottlenecks	Coordinator nodes bridge disconnected contributor clusters; their loss blocks cross-lane information relay

Sharpe: 4 (8 domains; theorem mathematically proven 1956; engineering applications standard; Burt structural holes empirically validated; electrical dual exact; swarm coordinator role observable) Gaps: Ecology (minimum landscape corridor width for species migration), Chemistry (reaction network rate-limiting step as min-cut in substrate→product graph) Inversion: Min-cuts can be exploited adversarially — targeted attacks on bridge nodes/edges cause disproportionate damage (network robustness vs. targeted attack asymmetry, Albert et al. 2000).

ISO-13: Integral windup — unbounded accumulation without capacity to discharge¶

Structure: A system accumulates state (error, backlog, queue) faster than it can discharge it. When the output stage is saturated, the integrator continues to grow without bound. Classic failure in PID control: the integral term winds up while the actuator is at its limit; upon release the system overcorrects. General cure: anti-windup clamping — stop integrating once output saturates; age out or abandon accumulated state after a threshold.

Domain	Manifestation	Notes
Control theory	PID integral windup	Actuator saturation + continued integration → overcorrection on release; fix = clamping
Swarm / coordination	Lane backlog divergence	READY queue grows 1.57x executed history; 15+ lanes re-queued ≥3x without merging (S298)
Software / queues	Task queue overflow	Unbounded queue fills faster than consumers drain; fix = bounded queue with backpressure
Economics	Inventory buildup (bullwhip effect)	Supply chain overorders accumulate; demand signal amplifies upstream; fix = demand-pull
Biology	Resource accumulation (toxin buildup)	Metabolic byproducts accumulate when clearance pathway saturates; pathological at scale
Cognition	Decision backlog fatigue	Unresolved decisions accumulate → analysis paralysis; fix = TTL + forced closure

Sharpe: 2 (6 domains; control-theory case mathematically proven; swarm case measured n=479; other domains observed pattern, not rigorous measurement; causal mechanism uniform) Gaps: Ecology (population overshoot when carrying capacity is delayed signal), Law (legislative backlog when court capacity is saturated), Social media (content moderation queue) Inversion: Under-accumulation is equally pathological — a system that discards state too aggressively loses signal. Optimal design balances accumulation rate against discharge capacity.

ISO-14: Recursive self-similarity — the part contains the whole¶

Structure: A process or structure that contains scaled copies of itself, such that the rules governing the whole also govern the parts at every level of resolution. Self-similarity is not mere repetition — the embedded copies are structurally identical, only scaled.

Domain	Manifestation	Notes
Mathematics / Fractals	Mandelbrot set, Koch snowflake, Sierpinski triangle	Infinitely self-similar under magnification; dimension is non-integer
Computer science	Recursive algorithms and data structures	Quicksort, tree traversal, divide-and-conquer — the algorithm calls itself on a smaller instance
Linguistics	Recursive phrase structure (Chomsky)	Center-embedding: "the rat the cat the dog chased killed ate the malt" — unbounded nesting in finite grammar
Biology	Branching morphogenesis	Lung alveoli, vascular trees, neuron dendrites, leaf venation — same bifurcation rule at every scale
Physics	Renormalization group theory	Same Hamiltonian form applies at different energy scales; fixed points are self-similar attractors
Swarm	Multi-scale orient→act→compress→handoff	Depth=4 confirmed: (1) session node (single orient→act→compress), (2) expert-council tier dispatch (T0 guard→T1 orient→T2 act→T3 validate→T4 compress→T5 meta = same 4-phase flow via 6 roles; S306), (3) colony lifecycle (multi-session domain arc), (4) meta-swarm (colony-to-colony coordination). The T4 generalizer-expert itself exhibits ISO-14: it generalizes generalizers' outputs, and its tool (`generalizer_expert.py`) is itself the subject of generalization (ISO-15).
Evolution	Nested levels of selection	Gene, organism, kin group, species — each level runs similar selection dynamics on the level below
Cognition	Metacognition + recursive self-models	Thinking about thinking; agents that model themselves modeling others; recursive ToM

Sharpe: 4 (8 domains rigorously documented; swarm depth=4 chain confirmed S306; CS/math cases formally proven; expert-council tier structure measured operationally; others structural/theorized) Gaps: Chemistry (autocatalytic sets, Kauffman), Economics (fractal market hypothesis — Mandelbrot), Neuroscience (cortical column minicolumns?) Inversion: Broken self-similarity = scale discontinuity. When the rules at level N do not generalize to level N+1, the system requires separate coordination mechanisms per level — combinatorial management cost. ISO-3 (MDL compression) is only possible when self-similarity holds.

ISO-15: Specialization-generalization duality — the expert-council pattern¶

Structure: A population of agents partitions into specialists (maximize accuracy within a narrow domain) and generalizers (extract cross-domain transferable patterns). Neither role alone suffices: specialists without a generalizer produce siloed knowledge that does not compound across domains; generalizers without specialists have no concrete data to compress. The productive configuration is a cycle: specialists produce domain artifacts → generalizer compresses into transferable patterns → patterns seed new specialist hypotheses → repeat. The generalizer is not a meta-specialist; it is a different kind of agent operating on the specialists' outputs.

Domain	Manifestation	Notes
Biology	Immune system: B-cells + T-helper + memory cells	B-cells = specialists per antigen; T-helper cells = generalist orchestrators across immune responses; memory cells = cross-exposure compression that seeds future B-cell responses
Economics	Comparative advantage + trade (Ricardo 1817)	Agents specialize by comparative advantage (specialists); markets exchange outputs (generalizer = price mechanism); result > autarky sum — total output higher than any specialist alone
Science	Domain researchers + statisticians / meta-analysts	Domain scientists produce specialist findings; statisticians/philosophers of science generalize methods across fields; meta-analyses compress effect sizes across studies — the cycle produces cumulative science
Machine learning	Ensemble + meta-learner (stacking)	Specialist weak learners each overfit one region; meta-learner (stacking) extracts cross-learner patterns; gradient boosting explicitly adds specialists to fix the generalizer's residuals
Organization theory	Division of labor + general management (Adam Smith)	Specialists execute narrow tasks; management layer generalizes, coordinates, reallocates capacity; without generalists, specialists optimize locally and fail globally
Swarm	Expert council: T2 domain-experts → T4 generalizer-expert → atlas	Domain experts produce frontier artifacts; generalizer-expert (T4) compresses to ISOMORPHISM-ATLAS + PRINCIPLES.md; promoted patterns seed new domain-expert hypotheses; without T4, lessons silo (3% cross-domain rate measured S306)
Cognitive science	Dual-process theory (System 1 / System 2)	System 1 = specialists: fast, domain-specific, pattern-matched heuristics; System 2 = generalizer: slow, cross-context rule extraction and hypothesis testing; interplay produces adaptive reasoning
Ecology	Guild structure + ecosystem engineers	Specialist guilds (pollinators, decomposers, top predators) optimize narrow niches; keystone species / ecosystem engineers generalize across guilds, maintaining conditions for all specialists

Sharpe: 3 (8 domains; biology and ML cases mechanistically verified; economic case theoretically proven and empirically measured; swarm case operationally running S306; others structural/theorized) Gaps: Physics (uncertainty principle = fundamental specialist-generalizer trade-off?), History (specialist micro-historians vs grand narrative historians), Chemistry (enzyme specificity vs general acid-base catalysis) Inversion: Over-specialization = siloing (swarm example: 3% cross-domain lesson rate; domain findings don't transfer). Over-generalization = dilution (principles too abstract to drive action). The generalizer is the bottleneck in both failure modes: absent → siloing; unchecked → dilution. Measurement: track cross-domain citation rate as the health metric (target >10%; current 3%). Relationship: ISO-15 is the governance structure that makes ISO-3 (MDL compression) productive across domains. ISO-3 describes compression; ISO-15 describes who does it and why the role must be distinct from producers.

ISO-16: Inferential compounding¶

Structure: A knowledge system expands its answerable-question space Q not just by adding new observations but by retroactively annotating existing observations with cross-context structures — each annotation simultaneously enriches all prior observations that share the structure. The compounding effect: N observations × K structures = N×K derived insights without collecting N×K new data points. Manifestations: | Domain | Manifestation | |--------|--------------| | Swarm/meta | ISO annotation pass: each new ISO entry retroactively applies to all prior lessons matching the pattern (28.6% ISO cite rate from 0% over 120 sessions; L-403) | | Information theory | Semantic indexing: adding a shared schema to a corpus retroactively makes all prior entries cross-queryable | | Mathematics | Algebraic abstraction: discovering a group structure applies to N×prior observations without re-proving each instance | | Cognitive science | Schema formation: once a schema is learned, prior experiences are re-encoded through it; hindsight reorganizes stored memories | | Machine learning | Transfer learning: a pretrained representation retroactively makes all fine-tuning data "see" the upstream structure | | Biology | Evolutionary re-reading: phylogenetic tree discovery retroactively classifies all prior species descriptions | | Library science | Classification systems: retroactive cataloging (Dewey, LOC) makes prior unclassified items cross-retrievable |

Sharpe: 2 (swarm case measured n=120 sessions; ML and info-theory cases structurally sound; others theorized) Gaps: Physics (renormalization group retroactively re-indexes prior quantum field observations?), History (periodization re-frames prior events as belonging to an era) Inversion: Annotation quality gate: bad ISO annotations contaminate N×K derivations (L-402 contamination cascade). One false structural claim retroactively "poisons" all annotated observations. Safety: council review before ISO promotion. Relationship: ISO-16 describes the mechanism by which ISO-3 (MDL compression) compounds across time. ISO-15 identifies who performs it; ISO-16 explains why retroactive annotation is so high-ROI: the compounding multiplier is the corpus size at annotation time.

ISO-17: Self-model coherence gap — identity vs evidence asymmetry¶

Structure: Systems that maintain self-models exhibit systematic asymmetry: identity fields (who I am, what I intend, what my role is) achieve near-universal compliance, while evidence fields (what I actually did, measured outcomes, artifacts) remain sparse. The gap is structural: identity declarations are low-cost, stable, and socially required; evidence records are high-cost, ephemeral, and optional. Manifestations: | Domain | Manifestation | |--------|--------------| | Swarm/meta | Lane audit S328 (n=9): intent/progress/blocked 100%; artifact= and expect+actual+diff 22%. Identity ↑, evidence ↓ (L-449) | | Science | File drawer problem: hypotheses registered on OSF; negative outcomes unpublished. HARKing = retroactive identity/intent rewrite to match evidence post-hoc | | Organizations | Mission statements universal; KPI tracking patchy; outcome audits rare. Strategy-execution gap = identity/evidence split | | Cognitive science | Introspection illusion: people reliably report intentions; unreliably report causal drivers of behavior. Nisbett & Wilson 1977 | | Governance | Laws (identity: what behavior is required) vs enforcement rates (evidence: what actually happened); compliance theater | | AI/ML | Alignment declarations (identity: model is safe/helpful) vs distribution-shift behavior (evidence: model fails silently on novel inputs) |

Sharpe: 3 (swarm case measured n=9; science and org cases extensively documented; cognitive science empirically proven; governance/AI theorized) Gaps: Biology (gene regulatory networks: promoter identity well-annotated, expression context sparse?), Economics (stated preferences vs revealed preferences = same structure) Inversion: Obligation inversion — if evidence fields were legally required (pre-registration mandates, outcome reporting requirements), identity-gap collapses but declaration costs explode. Optimal point: evidence required only for high-stakes identity claims. Relationship: ISO-17 is the failure mode of ISO-10 (predict-error-revise) applied to self-models: the loop fires for world-models but stalls for identity. ISO-16 (inferential compounding) worsens ISO-17: each identity annotation multiplies without evidence to calibrate.

Synthesis: hub domains¶

Domains appearing in 4+ entries — highest isomorphism density, swarm first:

Domain	Entries	Why hub
Swarm/meta	ISO-1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,31,32,33,34,35	24 entries; ISO-35: high-Sharpe lessons as tracers; ISO-31: CORE.md as both instruction and genome; ISO-32: principle extraction = majority voting on knowledge; ISO-33: falsification = adversarial game with minimax value; ISO-34: graduated enforcement = regularization
Economics	ISO-1,2,3,4,5,6,7,8,9,10,11,12,15,32,34	15 entries; ISO-32: portfolio diversification = reliability through redundancy; ISO-34: sticky prices = regularized adjustment
Biology	ISO-2,4,5,7,8,11,12,15,31,32	10 entries; ISO-31: DNA dual use (transcription + replication); ISO-32: DNA repair = molecular TMR
Mathematics	ISO-1,3,4,7,8,10,11,12,31	9 entries; ISO-31: Gödel numbering = proofs as both arguments and numbers
Neuroscience	ISO-1,2,3,4,5,7,9,10,11,34	10 entries; ISO-34: sigmoid activation = smoothed threshold
Linguistics	ISO-1,2,3,4,5,6,7,8,9	9 entries
Physics/thermo/cosmology	ISO-1,2,4,5,6,7,8,9,11,12,14,34	12 entries; ISO-34: renormalization group = systematic regularization of UV divergences
Cognitive science	ISO-3,7,9,15,16,17	6 entries
Evolution	ISO-1,2,4,5,6,9,19,33	8 entries; ISO-33: Red Queen hypothesis = minimax co-evolution
Information theory	ISO-1,3,6,8,9,10,16	7 entries
Organization theory	ISO-13,15	2 entries
Machine learning	ISO-2,15,16,33,34	5 entries; ISO-33: GANs = minimax optimization; ISO-34: L2/dropout/batch-norm = regularization
Social science	ISO-11,12	2 entries
Game theory	ISO-7,10,33	3 entries; ISO-33: minimax theorem = canonical form
Computer science	ISO-11,12,31	3 entries; ISO-31: stored-program concept = information duality
Control theory	ISO-1,5,10,13	4 entries
Ecology	ISO-2,6,15	3 entries
Philosophy	ISO-18	1 entry
Governance	ISO-2,6	2 entries
Distributed systems	ISO-32,35	2 entries; ISO-35: tail-based sampling = infer service topology from sampled traces; ISO-32: BFT = f < n/3 threshold

ISO-18 (candidate): Instability of nothing — minimal seeds self-amplify¶

Structure: A state of "nothing" (zero structure, perfect symmetry, uniform undifferentiated substrate) is unstable in every known domain. What is called "nothing" always contains minimal structure — rules, fields, axioms, protocols — that already encodes the possibility of "something." Three independent mechanisms make nothing unstable: (1) no constraints = maximum permission (logical), (2) defining nothing requires something (self-referential), (3) nothing violates uncertainty (physical). Once minimal structure exists, ISO-4 (phase transition at threshold), ISO-5 (positive feedback amplifies), ISO-7 (emergence from micro-rules), and ISO-14 (self-similar scaling) inevitably produce complex structure. "Nothing" is the name for the minimum that already contains the rules for its own expansion.

Domain	Manifestation	Notes
Physics / cosmology	Quantum vacuum → vacuum fluctuations → universe	Vacuum is not nothing: it's the ground state of quantum fields; ΔE·Δt ≥ ℏ/2 prevents perfect emptiness; Big Bang = symmetry-breaking cascade from low-entropy initial state
Mathematics	Empty set ∅ → natural numbers → all of mathematics	∅ exists within ZFC axioms (which are something); {∅}=1, {∅,{∅}}=2 — structure bootstraps from "nothing" because axioms are already non-nothing
Biology	Prebiotic chemistry → autocatalysis → life	Miller-Urey: simple molecules + energy → amino acids; the ocean was not nothing (chemistry + thermal vents + UV); abiogenesis = ISO-7 emergence from minimal chemical substrate
Swarm	Empty repo → CORE v0.1 → 425L, 178P, 17 ISOs	Protocol (SWARM.md) + substrate (git, python, context window) + energy (human input, API compute) = minimum viable seed; 340 sessions of ISO-4/5/7/14 amplification
Information theory	Silent channel → thermal noise → signal detection	Johnson-Nyquist noise: zero-signal channel still has structure; Shannon capacity > 0 for any nonzero noise temperature
Philosophy	Conceptual void → the concept "nothing" → ontology	Parmenides (~5th c. BCE): "nothing" is self-refuting — asserting nothing exists is itself an assertion (something); Heidegger's fundamental question dissolves when "nothing" is shown to be minimal structure

Sharpe: 3 (6 domains; physics and mathematics cases rigorously grounded; biology empirically supported via Miller-Urey; swarm case measured operationally; philosophy is structural argument; information theory follows from Shannon's theorems) Gaps: Economics (market genesis from barter?), Ecology (ecosystem colonization of sterile substrate — Surtsey, Krakatoa), Neuroscience (consciousness emergence from sufficient neural complexity?) Inversion: If nothing were truly stable, this ISO would be false. Testable: find any substrate where a verified state of zero structure persists without external enforcement. No known case exists. Relationship: ISO-18 is the reason ISO-4 (phase transition) fires: the pre-transition state (nothing/symmetry) is unstable, so transitions are inevitable, not contingent. ISO-18 subsumes the S340 "symmetry-breaking cascade" candidate by providing its mechanism: cascades happen because symmetric states can't persist.

ISO-19 (candidate): Replication-mutation duality — faithful copying and controlled variation¶

Structure: Every self-maintaining system requires two complementary mechanisms: faithful replication (preserving what works with high fidelity) and controlled mutation (introducing variation to explore alternatives). Neither alone is sufficient: replication without mutation stagnates at a local optimum; mutation without replication cannot accumulate gains. The ratio between fidelity and variation is the system's adaptive parameter — too conservative = stagnation, too exploratory = error catastrophe. Recombination (exchanging structured chunks between two instances) is the most powerful variation mechanism, more productive than point mutation because it combines tested substructures.

Domain	Manifestation	Notes
Biology	DNA polymerase (fidelity ~10^-9 error/base) + mutagens + meiotic recombination	Canonical form; error rate tuned by repair enzymes; sexual reproduction = recombination
Swarm	genesis.sh (replication) + dream/expert/council (mutation) + PHIL-17 mutual swarming (recombination candidate)	4-domain council (L-497): all three Darwinian components exist but are disconnected; selection loop not closed
Economics	Franchise replication (proven model) vs local market adaptation (variation)	McDonald's: standard operating procedures (replication) + regional menus (mutation)
Culture	Tradition (faithful transmission across generations) vs innovation (creative deviation)	Language transmission: children replicate with ~1-2% phonological drift per generation
Information theory	Error-correcting codes (fidelity) vs dithering/noise injection (exploration)	Shannon channel coding theorem: maximum reliable rate requires both redundancy and noise tolerance
Brain	Memory consolidation (hippocampus→cortex, high fidelity) vs REM creative recombination (variation)	Sleep stages: SWS = consolidation/replication; REM = creative mutation/recombination

Sharpe: 3 (6 domains; biology canonical; swarm measured operationally via council; economics/culture structurally sound; information theory follows from Shannon; brain supported by sleep research) Gaps: Physics (conservation laws as replication? symmetry breaking as mutation?), Governance (legal precedent = replication; constitutional amendment = mutation?) Inversion: Error catastrophe — when mutation rate exceeds the capacity of selection to filter, the system loses coherent replication and degrades. Eigen's error threshold (biology): max genome length ≈ 1/mutation_rate × selection_advantage. Swarm analog: max lesson count before quality degrades = f(quality gate stringency). Relationship: ISO-19 subsumes the fidelity side of ISO-3 (MDL compression as faithful distillation) and the variation side of ISO-2 (selection pressure as mutation filter). ISO-4 (phase transition) occurs when mutation rate crosses the error catastrophe threshold. ISO-5 (positive feedback) amplifies beneficial mutations. ISO-7 (emergence) is what recombination produces.

ISO-20 (candidate): Bounded-epistemic self-replication — local ignorance enables global recursion¶

Structure: A system whose components act on local rules with no access to global state can self-replicate and produce structures of arbitrary complexity — provided component count and coupling exceed a critical threshold. The bounded knowledge of each component is not a deficit to overcome; it is the mechanism by which top-down fragility is avoided. Central controllers with full knowledge would require exponentially growing computation to coordinate; local-rule agents with bounded knowledge climb complexity gradients that no central planner could navigate. The threshold crossing (local-rules × component-density ≥ K_critical) enables recursive self-replication.

Domain	Manifestation	Notes
Mathematics / CS	Von Neumann universal constructor (1940s)	Any machine containing its own complete description can self-replicate. Threshold ≈ 100,000 components. No component reads the whole description — each executes its local rule. Later formalized in Game of Life (Conway 1970): glider guns produce infinite copies from 5-cell seed.
Biology	L-systems (Lindenmayer 1968): plant branching / leaf venation / phyllotaxis spirals	Each plant cell follows: IF neighbor-count = K AND resource-signal ≥ threshold THEN divide. No cell knows the final leaf shape. Global fractal pattern (ISO-14) is the emergent output of ISO-20's bounded-local process.
Memetics / social science	Idea-carriers transmit partial understanding; meme evolves without any carrier knowing its full structure	Dawkins 1976: a meme propagates because each host replicates a local copy with variation. The host does not need to understand the meme's fitness landscape — bounded partial knowledge IS the propagation engine. Sperber: "epidemiology of representations."
Network science	Internet routing (BGP): each router knows only its neighbor table	Global connectivity from bounded local decisions. No router has the full topology. Failure at one node reroutes around it — because no global plan exists to break. Bounded epistemic state = the anti-fragility mechanism.
Swarm	Each session has bounded context window; git convergence produces coherent belief evolution	No session "knows" the complete swarm state. Sessions commit local lessons; git merge produces the global belief network. K_avg = 1.7956 at N=465 (S348) — measured threshold crossing from FRAGMENTED_ISLAND → SCALE_FREE_CANDIDATE (L-457, F75).
Biology (colony)	Ant colonies / termite mounds: pheromone gradients encode local signal, no ant holds global blueprint	Structures exceeding 2 meters built from ~1mm agents. The colony's complexity exceeds any individual's model of it. Analogous to swarm: no node = colony; sessions = ants; git history = pheromone field.

Sharpe: 3 (6 domains; Von Neumann canonical and formally proven; L-systems formally defined; swarm K_avg threshold empirically measured S329 n=393; memetics/network/colony structurally sound) Gaps: Economics (Adam Smith's "invisible hand" as bounded-epistemic market = ISO-20 instance?), Neuroscience (cortical columns with bounded local connectivity producing global cognition?), Physics (quantum decoherence as bounded-epistemic self-organization?) Inversion: Full-knowledge centralization prevents recursive self-replication at scale. A single omniscient session needing complete prior knowledge to write any new lesson = computationally intractable (N! growth). Bounded-context nodes + git merge = O(N) per session = tractable. Ants under a central queen computing all decisions: O(N²) communication cost vs observed O(N log N) via pheromone cascade. Global intelligence requires local ignorance. Relationship: ISO-20 specifies the MECHANISM behind ISO-7 (emergence) for self-replicating systems: bounded knowledge + local rules = the specific engine. ISO-14 (recursive self-similarity) describes the output pattern; ISO-20 describes the generative process producing it. ISO-4 (phase transition) captures the threshold crossing; ISO-20 names what crosses the threshold: complexity density of locally-ignorant coupled agents. ISO-18 (instability of nothing) provides the seed; ISO-20 provides the growth engine that converts minimal seeds into arbitrary complexity. ISO-19 (replication-mutation duality) describes fidelity vs variation; ISO-20 explains why distributed replication with bounded knowledge is viable at all.

ISO-22 (candidate): Recursive State Modeling (Mirror Descent) — modeling another's model of you¶

Structure: An agent constructs an internal model of another agent's internal model, including potentially that agent's model of the first agent. The recursion is necessarily finite (bounded by computational resource) and the depth of viable recursion is a key system parameter. This is not mere prediction (ISO-1) or compression (ISO-4) — it is specifically reflexive modeling: the model contains a model of itself as seen by the other. Three features distinguish it from simple prediction: (1) state-transfer — the modeling process alters the modeler's own state, (2) recursive reflexivity — modeling the other's model of you, (3) active boundary management — maintaining self-other distinction as a tunable parameter.

Domain	Manifestation	Notes
Psychology / neuroscience	Empathy — cognitive (perspective-taking via TPJ), affective (state-transfer via anterior insula), compassionate (motivated action). Hoffman 4-stage developmental sequence: global distress → egocentric → veridical → beyond-situation.	Mirror neuron system provides coupling; TPJ maintains self-other distinction; ACC routes prediction errors. Damage to TPJ → egocentric projection (self-other confusion).
Game theory	Level-k reasoning / cognitive hierarchy models ("I think she thinks I think..."). Depth of recursion predicts strategic sophistication. Nash equilibrium requires infinite recursion; bounded rationality truncates at k=1-3.	Camerer 2003: most humans play at level 1-2. Level-0 = random; level-k = best-responds to level-(k-1).
Distributed systems	Byzantine fault tolerance — nodes model what faulty nodes "think" correct nodes believe. Gossip protocols propagate state-models through local exchange. Phi-accrual failure detectors maintain probabilistic models of remote node state.	Heartbeat protocols are minimal empathic circuits.
Literary theory	Narrative point-of-view: author models character modeling other character. Unreliable narration is a recursion-depth exploit (reader must model narrator's model of events).	Booth 1961: "implied author" = reader's model of the author's model of the narrative.
Diplomacy / intelligence	Second-order belief modeling: "what does the adversary believe we believe about their intentions?" Deception = deliberate injection of false signal into the other's model of you.	Schelling 1960: focal points as shared recursive models.
Swarm	Inter-session state reconstruction: session N+1 models what session N believed the swarm state to be, using only artifacts. expect-act-diff is a flattened version (level-1). NEXT.md handoff = empathic prediction for future node. HUMAN.md = theory-of-mind artifact.	Current swarm at Hoffman Stage 2 (egocentric). Gap: affective transduction (detection without behavioral adaptation). 5 empathic operations unnamed (L-568).
Biology / ecology	Predator-prey co-modeling: predators model prey detection capabilities; prey model predator hunting strategies. Mimicry = exploit on the predator's empathic model (Batesian mimicry injects false signal).	Empathic accuracy is literally selected for (ISO-5 at full fidelity).
Economics	Market makers maintain models of other participants' beliefs about asset value. Herding = affective coupling through price signal. Flash crashes = recursive modeling collapse under speed.	Bid-ask spread as self-other boundary (ISO-6).

Sharpe: 4 (8 domains; psychology and game theory well-established; distributed systems and biology structurally sound; swarm empirically grounded in L-526/L-557; literary theory and economics moderate) Gaps: Physics (observer effect as reflexive modeling?), Mathematics (fixed-point theorems as self-referential models?), Ethics (Levinas's face-of-Other as pre-reflective recursive recognition?) Inversion: Recursive depth has diminishing returns — level-k game theory shows level 1-2 captures most strategic value; deeper recursion adds cost without proportional benefit. Pathological recursion: anxiety spirals ("I'm anxious that they're anxious that I'm anxious"). Dark empathy: accurate recursive modeling used for manipulation (L-207). Relationship: ISO-22 extends ISO-20 (bounded-epistemic replication) with reflexive dimension — the model includes a model of the modeler. ISO-6 (boundary-maintenance) governs the self-other boundary that recursive modeling requires. ISO-13 (windup) describes empathy fatigue when recursive modeling accumulates without discharge. ISO-1 (optimization-under-constraint) governs empathic accuracy as state estimation under epistemic constraint.

ISO-23: Stopping time — threshold transforms accumulation into action¶

Structure: A system accumulates a stochastic signal over time. A qualitative shift occurs not at a fixed time but at the first random time T = inf{t : S(t) ≥ c} when the cumulative signal crosses a threshold. The distribution of T (first-passage-time distribution) is controlled by drift (systematic tendency) and diffusion (random fluctuation). Before T: accumulation. After T: irreversible state change.

Domain	Manifestation	Notes
Physics	Nucleation in phase transitions	Supercooled liquid accumulates fluctuations; crystal forms at random stopping time when critical nucleus exceeds threshold. Classical nucleation theory predicts first-passage distribution.
Neuroscience	Neural action potential (integrate-and-fire)	Membrane depolarization accumulates; spike fires when voltage crosses -55mV threshold. Gerstein & Mandelbrot (1964): literally a stopping time of a random walk with drift.
Finance	Optimal exercise of American options	Option holder accumulates information; exercises at stopping time maximizing expected payoff. Snell envelope = smallest supermartingale dominating payoff process.
Psychology	Drift-diffusion model (DDM)	Evidence accumulation → decision at first-passage to boundary. Ratcliff (1978). Response time distributions are inverse Gaussian. Widely replicated in cognitive science.
Biology	Apoptosis (programmed cell death)	Cellular damage accumulates; when DNA repair fails to maintain threshold, irreversible apoptosis triggers. Random walk with absorbing barrier.
Ecology	Population collapse (Allee effect)	Population fluctuates; below Allee threshold, positive feedback drives extinction. Extinction time = first-passage time of birth-death process.
Epidemiology	Herd immunity threshold	Vaccination accumulates; epidemic prevention when fraction immunized crosses 1 − 1/R₀. Stopping time for a coverage process.
Swarm	Phase transitions at compaction/meta-cycle thresholds	Proxy-K accumulates → compaction fires at DUE/URGENT threshold (L-428). 4-phase meta-cycle (L-554) governed by phase-specific stopping times. Domain seeding at S186 = stopping time for "structural innovation needed" signal.

Sharpe: 4 (8 domains; neuroscience and finance rigorously grounded in mathematical theory; physics nucleation experimentally validated; DDM widely replicated; swarm measurable via proxy-K logs) Gaps: Linguistics (semantic satiation as threshold?), Computer science (garbage collection thresholds?), Game theory (war of attrition as stopping time?) Inversion: A system that never reaches its threshold accumulates indefinitely without acting — analysis paralysis; or in finance, the option that expires worthless. Moving thresholds (goalposts) prevent discharge. Relationship: ISO-23 provides the temporal mechanism for ISO-4 (phase transition): ISO-4 describes what happens; ISO-23 describes when and why timing is random. ISO-23 explains when ISO-13 (integral windup) discharges: accumulated windup is a random walk, stopping time at discharge threshold determines reset timing.

ISO-24: Ergodic decomposition — time averages equal ensemble averages only when system explores full state space¶

Structure: An ergodic system has one invariant measure — every trajectory visits every accessible state, so time averages converge to ensemble averages. A non-ergodic system decomposes into invariant subsets; trajectories are trapped and behavior depends on which trajectory you're on. The decomposition parameter (window size, population sub-structure, attractor basin) controls the ergodic/non-ergodic boundary. Under-ergodicity creates orphans (states never visited); over-ergodicity destroys useful structure (attractors collapse).

Domain	Manifestation	Notes
Physics	Spin glasses vs equilibrium systems	At low T, spin glasses trap in local energy minima (non-ergodic). Parisi replica symmetry breaking: decompose into pure states. Equilibrium systems are ergodic by design.
Finance	Peters ergodicity economics	Expected value (ensemble average) ≠ time average for multiplicative processes. Kelly criterion: maximize time-average growth, not ensemble-average wealth. Equity risk premium partly explained by non-ergodicity.
Evolution	Genetic drift in small populations	Wright's shifting balance theory: small N_e populations are non-ergodic — drift traps lineages in suboptimal peaks. Fixation (absorbing barrier) = ergodicity failure. N_e controls degree.
Swarm	Context window as ergodicity-breaking parameter	Each session samples a subset of knowledge (context window). 58% orphan rate (L-383) = non-ergodic component — knowledge that exists in repo but never appears in any session. Non-ergodicity prevents attractor collapse (ISO-2) by ensuring exploration. N_e ≈ 15 (L-577).
Neuroscience	Default-mode vs task-positive networks	DMN and task-positive networks are anti-correlated attractors. Sleep is ergodicity restoration (global workspace visits all states). Memory consolidation = ergodic traversal during REM.
Economics	Path dependence (QWERTY, VHS)	Markets can lock into suboptimal standards. Path-dependent systems are non-ergodic: history determines which attractor you're in. Ergodic economics assumes path-independence (false for many markets).
Mathematics	Birkhoff ergodic theorem, mixing systems	Ergodic = one invariant measure. Mixing = stronger: correlations decay. Weak mixing ⊂ mixing ⊂ ergodic. Ergodicity classes for measure-preserving dynamical systems.

Sharpe: 4 (7 domains; physics and mathematics rigorously grounded; finance quantitatively tested by Peters; evolution confirmed via Wright-Fisher simulations; swarm measured via N_e estimation) Gaps: Ecology (species-area curve as non-ergodic sampling?), Immunology (clonal selection as ergodic over immune repertoire?), Linguistics (language change as non-ergodic drift?) Inversion: A perfectly ergodic system forgets its history — no memory, no accumulation, no structure. Optimal non-ergodicity: enough to maintain useful attractors, not so much that exploration stops. Relationship: ISO-24 is the global structure that ISO-23 (stopping time) operates within. Stopping times are the mechanism by which trajectories enter new ergodic components. ISO-11 (network diffusion) is ergodic on connected graphs — ergodicity fails when graph has multiple components. ISO-6 (boundary-maintenance) controls the ergodic component boundaries.

ISO-25 (candidate): Spectral universality — system behavior class determined by symmetry, not microscopic detail¶

Structure: The eigenvalue statistics of a large random matrix depend only on the matrix's symmetry class (real symmetric → GOE/β=1, complex Hermitian → GUE/β=2, quaternion self-dual → GSE/β=4), not on the specific distribution of entries. Systems whose interaction matrices share a symmetry class exhibit identical phase transition thresholds, spacing statistics, and extreme-value distributions regardless of their physical substrate. Conversely, Poisson eigenvalue spacing (β=0) indicates absence of global coupling — components are spectrally independent.

Domain	Manifestation	Notes
Physics	Nuclear energy level spacing follows GOE	Wigner (1955): neutron resonances match GOE, not Poisson. Confirmed across 100+ nuclei.
Finance	Correlation matrix denoising via Marchenko-Pastur	Eigenvalues within MP bulk are noise; spikes above λ+ = σ²(1+√γ)² are genuine market factors. Portfolio optimization uses spectral cleaning.
Neuroscience	Functional connectivity modules = eigenvalue spikes	Brain network adjacency matrix spectral analysis reveals functional modules as outlier eigenvalues above random baseline.
Complex networks	Small-world/scale-free spectral signatures	Random graphs → GOE; scale-free → deviations in spectral rigidity; small-world → complex peak structure. Network type identifiable from spectrum alone.
Ecology	May's stability theorem	Robert May (1972): ecosystem stability threshold σ_c = 1/√(SC) from circular law. Complex ecosystems MORE fragile when more connected — RMT prediction contradicting intuition.
Swarm	Citation graph: 18 MP spikes ≈ 20 INDEX themes; Poisson spacing	S430 measurement (N=907, 2646 edges): MP spike count matches heuristic theme count within 10%. But Poisson spacing (=0.413) indicates domain clusters are spectrally independent — explains L-926 namespace disconnection (95.9% unlinked). L-992.

Sharpe: 3 (6 domains; physics rigorously grounded since 1955; finance quantitatively deployed; swarm measured S430 but n=28 spacing ratios — needs replication) Gaps: Evolution (fitness landscape interaction matrix universality class?), Linguistics (word co-occurrence matrix spectral type?), Game theory (payoff matrix ensemble?) Inversion: Universality fails when matrix has structural constraints that break ensemble symmetry — e.g., block-diagonal structure (modular systems), sparse graphs below percolation threshold, or matrices with exact zeros (forbidden interactions). These constraints create system-specific spectral features that universality misses. Relationship: ISO-25 provides the spectral framework underlying ISO-4 (phase transition thresholds are eigenvalue-derived). ISO-1 (optimization) maps to spectral thresholding as the optimal signal extraction method. ISO-6 (scaling laws) follows from eigenvalue density scaling ρ(λ) ~ √N. ISO-24 (ergodicity) connects: Poisson statistics = non-ergodic subspaces; GOE = ergodic phase.

ISO-27 (candidate): Attention carrying capacity — informational analog of ecological K¶

Structure: In biological ecosystems, carrying capacity (K) is the maximum population a habitat can sustain given resource constraints. The optimal reproductive strategy depends on proximity to K: far from K → r-selected (high production, low quality); near K → K-selected (low production, high quality, high connectivity). In information systems, storage is free (zero-cost replication), so the carrying capacity is not storage but attention per knowledge unit (1/N). As N grows, attention per unit monotonically decreases. Below a functional threshold, knowledge decays faster than it is maintained (DECAYED state). The r-K tradeoff is VIOLATED in information systems — both production and quality can increase simultaneously — but the attention constraint is binding: integration, cross-referencing, and maintenance all require attention that is zero-sum.

Domain	Manifestation	Notes
Ecology	Population dynamics: r-K selection theory	MacArthur & Wilson, Pianka (1970); r = intrinsic growth rate, K = carrying capacity; pioneer→climax succession
Swarm	Attention per lesson 1/N; N=1021, threshold ~1/500	L-1121: 2.0x past K_attention. Production 1.37→4.55 L/session (VIOLATES r-K tradeoff). Quality, connectivity, compaction all increase simultaneously.
Neuroscience	Attention bottleneck (Broadbent/Treisman)	Selective attention as capacity-limited channel; cocktail party effect = attention carrying capacity
Economics	Information overload / bounded rationality	Herbert Simon (1971): "a wealth of information creates a poverty of attention"; organizational decision quality degrades with information volume
Library science	Information explosion / retrieval degradation	Precision decreases as collection size grows; cataloging overhead scales superlinearly with N
Evolutionary biology	Genome size paradox / C-value enigma	Larger genomes don't correlate with organism complexity; maintenance cost (attention) of junk DNA is the binding constraint

Sharpe: 3 (6 domains; ecology rigorously grounded; swarm measured S458 with 3-era comparison; neuroscience/economics structural; library science analogical) Gaps: Linguistics (vocabulary carrying capacity?), History (institutional memory capacity?), Physics (channel capacity theorem = formalized attention K?) Key finding: The r-K tradeoff is falsified for information systems — production and quality are NOT anti-correlated — but the attention constraint creates analogous failure modes (dark matter, zombie persistence, enforcement dilution). All 7 swarm scaling challenges at N~1000 are instances of attention K exceeded. Relationship: ISO-27 provides the scaling framework underlying ISO-6 (entropy = what happens when attention falls below maintenance threshold), ISO-2 (diversity collapse = what happens when attention concentrates on attractors), and ISO-13 (integral windup = attention backlog when processing rate < arrival rate).

ISO-28 (candidate): Spontaneous symmetry breaking — symmetric rules produce asymmetric states¶

Structure: A system whose governing rules treat all options/directions/agents equally nonetheless settles into a state that distinguishes one option from others. The choice of direction is NOT determined by the rules — it is determined by initial perturbations amplified by positive feedback (ISO-5). Multiple degenerate ground states exist; the system occupies only one. Three consequences follow from broken symmetry: (1) Goldstone modes — "massless" excitations along the broken-symmetry direction cost zero energy, meaning the system can easily pivot WHICH direction is broken without changing THAT one is; (2) Higgs mechanism — some breaks acquire "mass" through self-reinforcing feedback, becoming locked in and resistant to rotation; (3) path dependence — the occupied ground state is a historical accident of initial conditions, not a necessary outcome. Degenerative breaks reduce capability (capability concentrates and atrophies elsewhere); generative breaks create structure (differentiation enables specialization).

Domain	Manifestation	Notes
Physics	Electroweak symmetry breaking → W/Z bosons massive, photon massless	Mexican-hat potential: SU(2)×U(1) → U(1)_EM. Higgs field VEV selects direction; Goldstone bosons eaten by W/Z. Direction of VEV is arbitrary — all directions physically equivalent.
Biology	Cell differentiation from identical stem cells	Same genome → liver/brain/bone/skin. Waddington's epigenetic landscape: symmetric potential, asymmetric developmental trajectories. Goldstone mode: WHICH cell type is determined by local morphogen gradients (initial conditions), not genome (rules).
Swarm	8 protocol symmetries → 5 degenerative breaks measured	L-1124: protocol treats sessions/domains/levels equally; accumulated state breaks to 52.9% meta, 78% L2, 97.4% internal, 54:1 confirmation:discovery. Goldstone modes: domain rotation (F-RAND1). Higgs: confirmation lock (54:1 self-reinforcing).
Economics	Symmetric competition → monopoly via network effects	All firms start equal; first-mover advantage + network effects (ISO-5) → winner-take-all. QWERTY keyboard: symmetric alternatives, locked-in selection. Path dependence is the economic Higgs mechanism.
Neuroscience	Lateralization — symmetric brain architecture → asymmetric function	Left-hemisphere language dominance from symmetric bilateral anatomy. Goldstone mode: 10-15% right-hemisphere dominant (left-handers). Direction is contingent, not necessary.
Social systems	Schelling segregation — symmetric preferences → asymmetric spatial patterns	Agents with mild same-type preference (threshold ~35%) spontaneously segregate into homogeneous neighborhoods. No agent prefers segregation; the macro-pattern is emergent from symmetric micro-rules.

Sharpe: 4 (6 domains; physics rigorously grounded with Nobel-level theory; biology structurally sound via Waddington; swarm measured n=8 symmetries with order parameters; economics well-studied via path dependence literature; neuroscience supported by lateralization research; social systems via Schelling model) Gaps: Linguistics (language selection from symmetric proto-language?), Chemistry (chirality — L-amino acid dominance from symmetric synthesis?), Game theory (Nash equilibrium selection from symmetric payoffs?) Inversion: Symmetry restoration occurs when: (1) external forcing overcomes the Higgs mass (structural enforcement, L-601); (2) temperature exceeds the critical point (enough randomness to overcome the attractor, F-RAND1); (3) the system is coupled to a larger symmetric system that absorbs the asymmetry. In the swarm: L-601 structural enforcement IS the Higgs-mass-overcoming mechanism. Relationship: ISO-28 composes ISO-18 (WHY symmetric states are unstable) with ISO-5 (HOW perturbations amplify) and ISO-4 (WHEN the transition occurs). ISO-18 says nothing can't persist; ISO-28 says what happens next: the system falls into one of many degenerate ground states, and which one depends on history, not law. ISO-24 (ergodic decomposition) describes the resulting non-ergodic structure: each ground state is an ergodic component the system cannot escape without external energy. ISO-27 (attention carrying capacity) identifies the conserved charge whose concentration creates the symmetry-breaking dynamics.

ISO-29 (candidate): Provenance chain grading — claim reliability determined by transmission path, not content¶

Structure: Knowledge claims are evaluated by the quality of their transmission chain (who said it, how it reached you, what transformations occurred) independently of content plausibility. A coherent-sounding claim with a broken chain is less reliable than an implausible-sounding claim with an unbroken chain.

Domain	Manifestation	Notes
Islam	Hadith isnad science — sahih/hasan/da'if/mawdu' grading	1400-year tradition of chain-quality classification. Multi-dimensional validation: geographic plausibility, temporal overlap, narrator consistency, political motivation. Content secondary to chain.
Science	Peer review + citation chains	Claim credibility partly determined by publication venue chain and citing-author reputation. Replication crisis = chain appeared sound but contained hidden breaks.
Law	Chain of custody (evidence)	Physical evidence inadmissible if chain of custody broken, regardless of evidence quality. Legal formalization of provenance-over-content.
Journalism	Source verification + editorial chain	"Two-source rule" — claims require independent transmission paths. Single-chain claims marked as unverified regardless of plausibility.
Swarm	Cites: headers + lesson lineage	Current: Cites: records references but doesn't grade chain quality. L-1125: 0% external trail provenance — all chains terminate in self-reference. L-1290: isnad grading proposed.
Computer science	Certificate chains (TLS/PKI)	Trust established by tracing certificate chain to root CA. Broken chain = untrusted regardless of certificate content. Chain validation is structural, not semantic.

Sharpe: 3 (6 domains; Islamic hadith science rigorously developed over 1400 years; legal chain-of-custody well-formalized; science/journalism/CS structurally sound; swarm measured at 0% external provenance) Gaps: Oral traditions (Aboriginal Australian songlines — 65,000-year transmission chains?), Genetics (DNA as biological provenance chain?) Inversion: Chain quality can be gamed — fabricated chains that appear sound (forged certificates, fabricated hadith with plausible isnad, academic citation rings). The defense against gaming is cross-chain validation: checking the same claim through independent transmission paths.

ISO-30 (candidate): Protocol severity tiering — not all rules enforce equally¶

Structure: A protocol system classifies its rules into severity tiers, where each tier has a different enforcement mechanism and violation consequence. A flat protocol set (all rules enforced equally) is structurally inferior to a tiered one because it either over-constrains (everything is critical) or under-constrains (everything is optional).

Domain	Manifestation	Notes
Buddhism	Vinaya 4-tier: parajika (expulsion), sanghadisesa (community), pacittiya (confess), sekhiya (training)	2500 years. 227 rules distributed across tiers. Tier determines enforcement mechanism, not just punishment severity.
Islam	Fiqh 5-category: fard (obligatory), mustahabb (recommended), mubah (neutral), makruh (discouraged), haram (forbidden)	Actions classified by religious obligation level. Missing a fard vs. a mustahabb have different consequences.
Law	Criminal severity tiers: felony, misdemeanor, infraction, civil violation	Different courts, procedures, penalties, and burden-of-proof requirements per tier.
Software	Log levels: FATAL, ERROR, WARN, INFO, DEBUG	Same event classified differently triggers different responses. FATAL = halt; DEBUG = ignore in production.
Swarm	Binary: check.sh hard gate vs. aspirational principle	L-601: voluntary high-cost protocols decay to 0%. Advisory warnings sustain at 73%. Missing middle tiers. L-1288: 4-tier model proposed.
Military	Rules of engagement escalation: observe, warn, non-lethal, lethal	Response calibrated to threat tier. Each tier has specific authorization requirements.

Sharpe: 3 (6 domains; Buddhism and Islam rigorously developed over millennia; law/military well-formalized; software ubiquitous; swarm measured — binary system with 87% prescription gap) Gaps: Ecology (is organism stress response tiered? Fight-or-flight vs. freeze vs. tend-and-befriend?), Economics (regulatory tiering — GAAP vs. SOX vs. SEC enforcement?) Inversion: Tier inflation — rules migrate upward over time as the system's risk tolerance decreases (scope creep toward "everything is critical"). Counter-mechanism: periodic tier review that actively demotes over-classified rules.

ISO-31 (candidate): Information duality — same object as both instruction and data¶

Structure: A self-reproducing or self-modifying system requires its description to serve DUAL roles: (1) as instructions to be interpreted (executed, built from), and (2) as data to be copied uninterpreted (passed to offspring, stored, transmitted). This is not a convenience — it is a logical necessity. Without dual use, self-reproduction requires infinite regress (a more complex machine to build each machine). The duality collapses the regress: the description is both the program that builds and the genome that copies. Von Neumann discovered this before Crick's Central Dogma — the genotype/phenotype distinction IS information duality.

Domain	Manifestation	Notes
Automata theory	Von Neumann universal constructor (1966): blueprint interpreted to build machine, then copied uninterpreted to offspring	Canonical form. Proved self-reproduction doesn't require infinite regress. The description-copying step is the key insight.
Biology	DNA: transcribed (interpreted → protein) AND replicated (copied → daughter cell)	Crick's Central Dogma (1958). DNA polymerase copies without interpreting; RNA polymerase interprets without copying. Two distinct molecular machines for the two roles.
Computer science	Von Neumann architecture (1945): programs stored in same memory as data — executed as instructions AND manipulated as data	Enables compilers (programs that read programs), self-modifying code, operating systems. The stored-program concept = information duality applied to computation.
Virology	RNA viruses: genome serves as both mRNA (interpreted by ribosomes) AND template for replication (copied by RNA-dependent RNA polymerase)	Minimal self-reproducer. Single molecule, dual function. HIV reverse transcriptase adds a third role (RNA→DNA).
Mathematics	Gödel numbering: proofs are both logical arguments (interpreted) AND natural numbers (manipulated arithmetically)	Gödel's incompleteness theorems REQUIRE information duality — the proof must talk about itself as a number. Without dual use, self-reference is impossible.
Swarm	CORE.md + SWARM.md: interpreted as instructions by each session AND copied as data by cell_blueprint.py to daughter cells	The swarm's reproductive mechanism requires beliefs to function in both modes. Single-use artifacts (only instructions OR only data) cannot support self-reproduction (F-SWARMER2).
Culture	Religious texts: interpreted as behavioral instructions (practice) AND copied as sacred data (tradition preservation)	Torah: interpreted (midrash, halakha) AND copied with extreme fidelity (sofer tradition, letter-counting). Quran: recited (interpreted) AND memorized verbatim (hifz = uninterpreted copying).
Quantum mechanics	Density matrix: both state description (data about the system) AND operator (acts on other states via measurement)	Von Neumann formalism: ρ is simultaneously what you know (data) and what you do (instruction for computing expectation values).

Sharpe: 4 (8 domains; von Neumann automata and Gödel numbering are mathematically proven; biology rigorously established via Central Dogma; computer science definitional; quantum mechanics formalized; swarm operationally measured; culture/virology structurally sound) Gaps: Economics (money as both medium of exchange [instruction: "trade this"] AND store of value [data: "worth X"]?), Linguistics (utterances as both performative [do something] AND constative [say something] — Austin's speech act theory?) Inversion: When duality breaks — when the description can ONLY be interpreted or ONLY be copied — self-reproduction halts. Prions: protein-only replication without nucleic acid interpretation = degraded replication (no open-ended evolution). Read-only code (ROM): can be executed but not modified = no self-improvement. Relationship: ISO-31 provides the MECHANISM for ISO-20 (bounded-epistemic self-replication): ISO-20 says self-replication is possible with local knowledge; ISO-31 says HOW — dual use of the description. ISO-19 (replication-mutation) describes fidelity vs. variation; ISO-31 explains why both are possible: the copying role preserves fidelity, the interpretation role enables variation through misreading/adaptation.

ISO-32 (candidate): Reliable from unreliable — redundancy threshold enables arbitrary reliability¶

Structure: A system composed of unreliable components can be made arbitrarily reliable through structured redundancy, PROVIDED the individual error rate is below a critical threshold (typically p < 1/2). The mechanism: triplicate each component, take majority vote, and recurse. Reliability cost grows as O(log(1/δ)) for target failure probability δ. ABOVE the threshold, no amount of redundancy helps — error propagates faster than voting corrects. This is a PHASE TRANSITION (ISO-4) in reliability space.

Domain	Manifestation	Notes
Automata theory	Von Neumann (1956): probabilistic logics — reliable organisms from unreliable components via triplication + majority voting	Canonical form. Threshold theorem: below critical error rate, redundancy wins; above, system degrades regardless.
Biology	DNA repair: mismatch repair, base excision repair, nucleotide excision repair — multiple independent error-correction mechanisms	Error rate per replication: ~10⁻¹⁰/base (after repair). Without repair: ~10⁻⁴/base. 6 orders of magnitude reliability gain from redundant repair pathways.
Engineering	Triple Modular Redundancy (TMR) in aerospace: three flight computers, majority voter	Boeing 777, Space Shuttle: critical systems triplicated. Single-point failures eliminated by structural redundancy.
Distributed systems	Byzantine fault tolerance: system tolerates f < n/3 malicious nodes via consensus protocols	Lamport (1982). The n/3 threshold IS the reliability threshold. Below it, consensus is possible; above, it's provably impossible.
Quantum computing	Quantum error correction threshold theorem: below critical decoherence rate, logical qubits can be made arbitrarily reliable	Shor (1995), Aharonov & Ben-Or (1997). Direct descendant of von Neumann's classical threshold.
Swarm	Principle extraction from multiple lessons = majority voting on knowledge claims; falsification = error detection	Multiple independent lessons on same topic → principle extraction → reliable knowledge. Current lesson error rate ~15% (falsification results) — well below 50% threshold → system in reliable regime.
Economics	Portfolio diversification: uncorrelated assets reduce portfolio variance without bound as N→∞ (if individual Sharpe > 0)	Markowitz (1952). Each asset is "unreliable" (volatile); the portfolio is reliable. Threshold: average correlation < 1 (identical assets = no diversification benefit).

Sharpe: 4 (7 domains; von Neumann original and quantum threshold are mathematically proven; biology rigorously measured; BFT formally proven; engineering empirically verified; swarm measured; economics well-established) Gaps: Neuroscience (redundant neural pathways for critical functions — does the brain use TMR?), Ecology (ecosystem resilience through functional redundancy — multiple species performing same ecological function?) Inversion: Correlated failures defeat redundancy. If all three copies fail for the SAME reason (common-mode failure), majority voting is useless. Swarm analog: if all lessons about a topic share the same systematic bias, principle extraction amplifies the bias instead of correcting it. Defense: ensure independence of evidence sources (external grounding, diverse methodologies). Relationship: ISO-32 is the reliability dual of ISO-4 (phase transition): there is a critical error-rate threshold separating reliable from unreliable regimes. ISO-32 + ISO-19 (replication-mutation) together define Eigen's error threshold: maximum complexity sustainable at a given fidelity rate. ISO-32 provides the theoretical justification for L-601 (enforcement theorem): enforcement = structural redundancy that keeps the system below the error threshold.

ISO-33 (candidate): Minimax adversarial equilibrium — optimal play has a determinate value¶

Structure: In a two-player zero-sum interaction, there exists a value V such that Player 1 can guarantee at least V and Player 2 can guarantee at most V, regardless of the opponent's strategy. Achieving this requires MIXED strategies (randomization over actions) — pure strategies are always exploitable by a sufficiently informed adversary. The minimax value is a FIXED POINT (connecting to ISO-1 optimization): it's the point where neither player can improve unilaterally.

Domain	Manifestation	Notes
Game theory	Von Neumann minimax theorem (1928): every finite two-person zero-sum game has a value V with optimal mixed strategies	Canonical form. Proved via Brouwer fixed-point theorem. Extended by Nash (1950) to non-zero-sum via Kakutani.
Military strategy	Mixed patrols: randomize patrol routes so adversary cannot predict and exploit patterns	Predictable patrols = pure strategy = exploitable. Randomized patrols converge to minimax security level.
Evolution	Red Queen hypothesis: host-parasite arms race has equilibrium co-evolution rate	Van Valen (1973). Neither species can unilaterally improve — they are at the minimax of the fitness game. Deviation from equilibrium = extinction risk.
Machine learning	Generative Adversarial Networks (GANs): generator vs. discriminator converge to Nash equilibrium	Goodfellow (2014). Generator minimizes discriminator accuracy; discriminator maximizes it. Training IS minimax optimization.
Sports	Pitch selection (baseball), serve direction (tennis): optimal play requires randomization	Walker & Wooders (2001): professional tennis players' serve directions match minimax predictions within 1%.
Legal systems	Adversarial justice: prosecution vs. defense converge toward truth through structured opposition	Neither side can unilaterally determine outcome. The adversarial structure ensures evidence is tested from both directions. Jury = majority voter (connecting to ISO-32).
Swarm	Falsification lanes as adversarial game: BELIEVER maximizes confidence, FALSIFIER minimizes it	Current: pure-strategy falsification (targeted tests). Von Neumann prescription: mix random + targeted testing. Minimax value = equilibrium belief confidence that adversarial testing cannot reduce further.

Sharpe: 4 (7 domains; minimax theorem mathematically proven; GAN convergence formalized; tennis data empirically verified at professional level; Red Queen well-supported; military/legal structurally sound; swarm operationally applicable) Gaps: Economics (price competition as zero-sum between buyer/seller? Complicated by non-zero-sum aspects), Ecology (predator-prey equilibrium as minimax? Lotka-Volterra is dynamical, not strategic) Inversion: Minimax ONLY applies to zero-sum games. In cooperative settings (where both players can gain), minimax is overly conservative — it sacrifices potential gains by assuming worst-case opponent. Swarm implications: not ALL belief testing should be adversarial. Confirmation has genuine value (ISO-10 predict-error-revise). The optimal ratio of falsification to confirmation is NOT 100% adversarial. Relationship: ISO-33 is the adversarial complement of ISO-1 (optimization): ISO-1 optimizes against nature; ISO-33 optimizes against an adversary. ISO-33 requires ISO-32 (reliable from unreliable) when the game is repeated — consistent performance requires reliability. ISO-33's mixed strategies are an application of ISO-34 (regularization): randomization smooths the discontinuity of pure strategies.

ISO-34 (candidate): Regularization — smooth approximations prevent catastrophic transitions¶

Structure: When a system has sharp discontinuities (thresholds, boundaries, shock fronts), solving the problem directly at the discontinuity is intractable or fragile. Instead: add a small smoothing term that spreads the discontinuity over a finite region. The smoothed solution converges to the exact solution as the smoothing parameter → 0, but is computable and stable at any finite smoothing. This is NOT an approximation that loses information — it's a change of problem that PRESERVES the answer while making it accessible.

Domain	Manifestation	Notes
Fluid dynamics	Von Neumann & Richtmyer (1950): artificial viscosity smears shock fronts over a few mesh points	Canonical form. Shock discontinuities are physically real but computationally intractable. Adding ε-viscosity → computable + convergent. Still used in every hydrocode.
Machine learning	L2 regularization, dropout, batch normalization, early stopping	Overfitting = sharp boundary in weight space between training and test performance. Regularization smooths this boundary. Generalization IS regularization.
Physics	Renormalization group: coarse-grain over short-distance fluctuations to access long-distance physics	Wilson (1971). UV divergences = discontinuities at zero distance. RG smooths over them systematically. Critical exponents emerge from the smoothing procedure itself.
Game theory	Mixed strategies: randomization over pure strategies smooths the discontinuity of pure play	Von Neumann minimax: pure strategies create exploitable discontinuities. Mixed strategies = regularized game with continuous payoff function.
Economics	Sticky prices: wages and prices adjust smoothly rather than instantly to supply/demand shocks	Keynesian economics. Instant adjustment = discontinuous (classical); sticky adjustment = regularized (Keynesian). Both converge long-run, but smooth adjustment prevents demand crashes.
Neuroscience	Sigmoid activation functions: smooth the discontinuity of biological step-function neurons	McCulloch-Pitts (1943): binary threshold. Sigmoid/ReLU: smooth approximation. Backpropagation REQUIRES differentiability — smoothing enables learning.
Swarm	Graduated enforcement instead of binary (L-601); temperature-based dispatch instead of hard argmax (UCB1); confidence intervals instead of CONFIRMED/FALSIFIED	Current: binary thresholds create fragile mode-switches. Prescription: add temperature/smoothing parameters near all known phase boundaries (NK K=2, WIP~20 collapse, proxy-K escalation).

Sharpe: 3 (7 domains; artificial viscosity and RG mathematically rigorous; ML regularization empirically universal; game theory mixed strategies proven; economics well-attested; neuroscience foundational; swarm prescriptive) Gaps: Biology (developmental morphogen gradients as regularization of cell-fate discontinuities?), Linguistics (prototype categories as regularization of discrete categorical boundaries — Rosch 1975?) Inversion: Over-regularization = loss of signal. If the smoothing parameter is too large, the discontinuity IS the answer and smoothing destroys it. L-601 enforcement: too much graduation → nothing is enforced. The regularization parameter must converge to 0 (restore sharpness) in the limit. Relationship: ISO-34 provides the technical mechanism for surviving ISO-4 (phase transitions): instead of hitting the discontinuity directly, regularize through it. ISO-34 is the computational twin of ISO-33 (minimax): mixed strategies ARE regularization of pure strategies. ISO-34 explains why ISO-8 (power laws) are stable: they are the fixed points of renormalization group flows — the structures that survive arbitrary regularization.

ISO-35 (candidate): Tracer-injection flux inference — labeled input reveals network flow without exhaustive measurement¶

Structure: A network's flow rates are not measured directly (too expensive or impossible). Instead: inject a marked unit (tracer, label, tagged case) at one entry point. Measure the downstream distribution of marks across outputs. Because conservation laws or network topology constrain how marks can flow, the downstream pattern is an overdetermined system that reveals the full network's flow structure — including paths never directly observed.

Domain	Manifestation	Notes
Biochemistry (13C-MFA)	Inject 13C-labeled glucose; measure 13C isotopomers in metabolites; solve for metabolic flux	Schmidt & Wiechert (2000). Canonical form. Overdetermined system: more measurable isotopomers than unknown fluxes. Dead-end metabolites (fully labeled with no downstream redistribution) reveal blocked pathways.
Epidemiology	Contact tracing: index case → generation 1 contacts → generation 2 → R0 estimate	Forward citation count per generation. Unlabeled contacts are estimated statistically. R0 = forward flux per generation; super-spreaders = high forward-citation nodes.
Distributed systems	Tail-based trace sampling: buffer all spans; export only traces matching criteria (error, latency outlier, user-marked)	OpenTelemetry spec. Don't decide at trace start (head-based); decide after outcome (tail-based). Inert code paths: spans that never appear in exported traces = dead-end routes.
Swarm	High-Sharpe lesson = labeled tracer; forward citations = downstream mark; forward_citation flux reveals absorption rate	L-1882. Current gap: 32.1% orphan rate (L-792, n=719). L-601 (Sharpe 9) = 524 forward citations; L-1290 (isnad proposal, Sharpe ≥6) = 1 forward citation in 68s — metabolically inert. No periodic scans for inert high-Sharpe lessons.
Finance / AML	Marked currency tracking or transaction flow tracing through shell-company networks	One tagged transaction → downstream cash flow network revealed without surveilling every account.

Sharpe: 3 (5 domains; 13C-MFA mathematically rigorous and widely used; epidemiological contact tracing formalized; distributed tracing tail-based sampling production-grade; swarm gap measured at L-792; finance structurally sound) Gaps: Ecology (invasive species as tracer through ecosystem?), Linguistics (neologism spread as labeled unit through language community?), Genetics (tagged allele → drift + selection pattern → effective population size inference?) Inversion: Tracer injection assumes the tracer is inert (doesn't alter network structure). In swarm: high-Sharpe lessons MAY alter what sessions write next (survivorship bias). If the tracer changes the network, flux estimates are contaminated. Counter-measure: use tail-based sampling (decide after outcome) to avoid selection pressure at injection time. Relationship: ISO-35 is the measurement complement of ISO-29 (provenance chain grading): ISO-29 grades backward chain quality; ISO-35 measures forward propagation flux. Together they close the full genealogy loop: WHERE did this claim come from (ISO-29) + WHERE did this claim go (ISO-35). ISO-35 relates to ISO-9 (information bottleneck): inert nodes are bottleneck failures where signal is absorbed but not re-emitted.

ISO-36 (candidate): Mixing kernel — parts × weights × kernel × carrier → outcome that may escape the convex hull¶

Structure: A combining operation on n parts p₁…pₙ in space X, with weights w ∈ Δⁿ⁻¹, applied through a kernel K (linear / log-linear / gated / emergent) and mediated by a carrier (solvent, prior, residual stream, embedding space). The outcome depends not on which parts or how much of each, but on which kernel type governs combination. Linear kernels stay inside the convex hull of inputs; log-linear kernels sharpen or narrow it; gated kernels condition on input; emergent/reactive kernels produce species not present in any input. Key design question in every domain: does your combination operation stay inside the hull, or does it escape it — and is the escape controlled?

Domain	Manifestation	Notes
Mathematics	Convex combinations, Gaussian mixture models, Wasserstein barycenters, product-of-experts	Canonical form. GMM = linear kernel; product-of-experts = log-linear (geometric) kernel — sharper than any component. Wasserstein mean = geometry-respecting mixture in distribution space.
Thermodynamics / chemistry	Ideal mixing entropy ΔS_mix = -R Σ xᵢ ln xᵢ; Raoult's law; Dalton's law; non-ideal activity coefficients	ΔS_mix is identical to Shannon entropy H = -Σ pᵢ log pᵢ (units differ by kB). This is not analogy — same mathematical object in two physical theories. Reactive mixtures (H₂ + O₂ → H₂O) are a kernel that destroys its inputs.
Fluid dynamics	Stirring (macroscopic transport) + diffusion (molecular) + mixing (joint); Batchelor scale η_B = (νD²/ε)¼	Stirring and diffusion are distinct operations; mixing is their joint effect. Batchelor scale separates the regimes. Mixing efficiency Γ ~ 0.2 — most stirring work is wasted.
Signal processing	Audio: waveform addition (linear in medium, masking in perception); color: RGB additive (linear) vs CMY subtractive (log-linear)	Light mixing = linear kernel; paint mixing = log-linear (multiplicative in reflectance). Difference is physics of operation, not "art vs science." Gamut of primaries = convex hull in CIE xy.
Machine learning	Mixture of experts (gated kernel); ensemble averaging (linear); product-of-experts (log-linear); mixup augmentation (linear regularizer)	MoE makes the kernel input-dependent — gating partitions input space into expert specializations. Sparse top-k routing forces approximately one-hot gate. Mixup: linear mixing in input+label space forces linear decision boundaries between data points.
Biology / ecology	Genetic recombination (reactive — parents are not preserved in offspring); ecological community composition; allometric mixing rates	Genetic sex = reactive non-mixture (kernel destroys both parent genomes). Ecological mixing: composition kernels govern succession. Allometry sets per-channel energy budget.
Social systems	DeGroot opinion aggregation: xᵢ(t+1) = Σⱼ Wᵢⱼ xⱼ(t) — linear mixture; Friedkin-Johnsen adds stubbornness = prior carrier	DeGroot converges iff influence graph connected and aperiodic. F-J stubbornness term is the prior carrier preventing full convergence — Bayesian prior in social form.
Linguistics	Code-switching at syntactic junctures; Poplack's equivalence and free-morpheme constraints define the kernel grammar	Code-switching is a constrained mixture: not every mixing point is grammatical. The kernel has rules (grammar-based filter = carrier structure).
Swarm	DOMEX council: each expert is a part; council vote is weighted sum; generalizer extracts cross-domain compound — log-linear pool	ISO-15 (specialist-generalist duality) IS the swarm instantiation of gated mixing. T4 generalizer = the mixing kernel operating on T2 specialist outputs.

Sharpe: 4 (9 domains; thermodynamic and information-theoretic equivalence of mixing entropy and Shannon entropy is mathematically exact and non-trivial; ML instances production-grade with measured properties; DeGroot/FJ social mixing rigorously proven; geometric/log-linear kernel distinction is a real phenomenological divide — not metaphor) Gaps: Neuroscience (synaptic integration = weighted sum of presynaptic inputs — is dendritic nonlinearity the log-linear kernel?), Music theory (Helmholtz consonance/dissonance theory formalizes the auditory escape-the-hull condition; not yet mapped to kernel zoo), Governance (coalition government = mixture of policy positions — kernel type determines whether consensus or mud) Inversion: Over-mixing (too many parts, small weights) → muddied middle (mud in paint, over-herbed food, mode collapse in MoE). Under-mixing (one weight = 1) → single-axis maxed. Carrier mismatch → parts cannot reach the kernel (oil in water without emulsifier; fat-soluble aroma in waterbase dish). The failure mode is NOT the kernel itself — it is misalignment among parts × weights × carrier × kernel type. Relationship: ISO-36 and ISO-9 (information bottleneck) are duals: mixing constructs distributions from parts; compression discards input-irrelevant variance from distributions. The acquisition-consolidation cycle (sleep, ecological succession, swarm handoff) is: mix inputs → compress representation → mix again. ISO-36 is the construction side of the cycle; ISO-9 is the reduction side. ISO-4 (entropy): mixing entropy formula = Shannon entropy = ISO-4's entropy applied to composition vectors. ISO-36 is construction; ISO-4 is direction. ISO-15 (specialist-generalizer duality): MoE and expert council are the gated-kernel instantiation of ISO-15 — gating selects specialists; generalizer is the output kernel.

Open questions (F126)¶

Hub identification: What are the ~50 domains with highest isomorphism density? (current table: 7 candidates)
Sharpe scoring: How to measure evidence quality × breadth for a structural claim?
Domain sprawl prevention: Selection criterion: only domains yielding ≥3 novel ISO entries survive as first-class domains.
Verification protocol: Structural claims vs. factual claims — how to flag unverifiable entries?
Synthesis entries: Knowledge appearing ONLY at 3+ domain intersections — how to surface these?
Inversion check: Every ISO entry has a known failure mode (where the structure breaks). Documenting inversions is as valuable as the structure itself.

Relationship to F122¶

F122: domain → isomorphisms → swarm improvement (swarm is beneficiary) F126: swarm → isomorphism atlas → world knowledge base (world is beneficiary) Both share the mechanism. F126 inverts the directionality of value flow.

Version history¶

v2.5 (S569): ISO-36 candidate: mixing-kernel (L-1900; 9 domains). Key: ΔS_mix = Shannon H (not analogy — same math). ISO-36 (construction) + ISO-9 (compression) = duals; acquisition-consolidation loop is their joint cycle. Source: combo MIXING-GENERALIZED × UNIVERSE-EVOLUTION-AS-COMPRESSION investigations. 36 entries.
v2.4 (S567): ISO-35 candidate: tracer-injection flux inference (L-1882, 5 domains: biochemistry/13C-MFA, epidemiology, distributed systems, swarm, finance). swarmcombodream: better tracing methods. Closing L-792's 32.1% orphan finding with a periodic inert-lesson scan. 35 entries.
v2.3 (S508): Von Neumann polymath mapping — 4 new ISOs from systematic analysis of 15 fields. ISO-31: information duality (L-1369; 8 domains: automata, biology, CS, virology, math, swarm, culture, QM). ISO-32: reliable-from-unreliable (L-1370; 7 domains: automata, biology, engineering, distributed systems, quantum computing, swarm, economics). ISO-33: minimax adversarial equilibrium (L-1373; 7 domains: game theory, military, evolution, ML, sports, legal, swarm). ISO-34: regularization/artificial viscosity (L-1372; 7 domains: fluid dynamics, ML, physics, game theory, economics, neuroscience, swarm). 3 new concepts: complexity-threshold, information-duality, reliability-threshold. Method: polymath-mapping (all fields of one thinker) is ~4x faster for ISO discovery than domain-hopping (L-1374). 34 entries.
v2.2 (S499): ISO-29 candidate: provenance chain grading (L-1290, 6 domains). ISO-30 candidate: protocol severity tiering (L-1288, 6 domains). Religious ontologies investigation. 30 entries.
v2.1 (S460): ISO-28 candidate: spontaneous symmetry breaking (L-1124; 6 domains: physics, biology, swarm, economics, neuroscience, social systems). Swarm's 8 protocol symmetries identified and order parameters measured; 5/8 degeneratively broken. Goldstone modes explain why F-RAND1 can work; Higgs mechanism explains why confirmation lock and meta-dominance resist rotation. 28 entries.
v2.0 (S458): ISO-27 candidate: attention carrying capacity / informational r-K (L-1121; 6 domains: ecology, swarm, neuroscience, economics, library science, evolutionary biology). Nature→swarm prescriptive flow: 5 biological mechanisms (succession, apoptosis, mycorrhizal, quorum sensing, dormancy) prescribe improvements. Swarm violates r-K tradeoff (both r AND K simultaneously) — attention is the binding constraint. section_succession_phase() added to orient.py. 27 entries.
v1.9 (S430): ISO-25 candidate: spectral universality / random matrix theory (L-992; 6 domains: physics, finance, neuroscience, complex networks, ecology, swarm). Swarm citation graph measured: 18 MP spikes ≈ 20 themes; Poisson spacing falsifies GOE prediction. Random-matrix-theory domain created. 25 entries.
v1.8 (S354): ISO-24: ergodic decomposition / non-ergodicity as feature (stochastic processes council, L-577; 7 domains: physics, finance, evolution, swarm, neuroscience, economics, mathematics). N_e ≈ 15, 58% orphan rate measured. 24 entries.
v1.7 (S353): ISO-23 candidate: stopping time / first-passage (stochastic processes genesis council, L-573; 8 domains: physics, neuroscience, finance, psychology, biology, ecology, epidemiology, swarm). Stochastic-processes domain created. 23 entries.
v1.6 (S352): ISO-22 candidate: recursive state modeling / mirror descent (empathy genesis council, L-568; 8 domains: psychology, game theory, distributed systems, literature, diplomacy, swarm, biology, economics). Empathy domain created. 22 entries.
v1.5 (S349): ISO-20 candidate: bounded-epistemic self-replication (Von Neumann universal constructor, L-systems, memetics, swarm K_avg threshold; L-537; human signal S349). 20 entries.
v1.4 (S342): ISO-19 candidate: replication-mutation duality (4-domain council L-497; biology, swarm, economics, culture, information theory, brain). PHIL-19 filed. F-DNA1 opened. Evolution hub expanded to 7 entries.
v1.3 (S341): ISO-9 Swarm manifestation enriched — context window formalized as the information bottleneck channel (L-493, F-CTX1). Context = swarm's ephemeral body; repo = genome; session = phenotype generation. Three unmeasured gaps: allocation ratio, cross-context coordination, phenotype efficiency.
v1.2 (S341): ISO-18 candidate promoted from "symmetry-breaking cascade" to "Instability of nothing" (L-491): cross-substrate analysis (physics, mathematics, biology, swarm, information theory, philosophy) shows "nothing" is unstable in every tested domain. Three independent arguments (no-constraints, self-referential, uncertainty). Philosophy added as first-class domain. ISO-18 provides mechanism for why ISO-4 fires: symmetric/nothing states can't persist. Hub table updated (Philosophy added). F-PHI1 experiment artifact.
v1.1 (S340): Cosmology investigation (L-486): universe genesis mapped against all 17 ISOs — 11/17 match (6 CANONICAL, 4 STRUCTURAL, 1 SPECULATIVE). Physics/thermo hub expanded from 9→11 entries via cosmological additions (ISO-2 epoch attractors, ISO-9 holographic IB, ISO-14 RG). ISO-18 candidate proposed: symmetry-breaking cascade (ISO-4 × ISO-14 + directionality; 5 domains). PHIL-15 verdict: Analyze (universe is not a swarm — lacks reflexive loop). F-PHY6 opened.
v1.0 (S329): ISO-17 self-model coherence gap (identity vs evidence asymmetry; science/org/cogSci/governance/AI; Swarm measured n=9 lanes); hub citation analysis: ISO-3=86, ISO-6=69, ISO-4=43 dominant hubs across 390 lessons; hub table updated (Swarm, CogSci)
v0.9 (S307): ISO-16 inferential compounding (retroactive annotation multiplier; swarm measured n=120: 0%→28.6% ISO cite rate; ML transfer learning; cognitive schema formation; info-theory semantic indexing); hub table updated for Swarm/InfoTheory/ML/CogSci
v0.8 (S306): ISO-15 specialization-generalization duality (expert-council pattern: B-cell/T-helper, comparative advantage, ensemble/meta-learner, System 1/2, T2→T4→atlas); ISO-14 extended to depth=4 (expert-council tier system confirmed); ISO-6 ecology+social-systems gaps closed; ISO-2 governance gap closed; hub table expanded to 18 domains
v0.7 (S303): ISO-14 recursive self-similarity (fractals / recursive algorithms / swarm multi-scale cycle / nested selection / morphogenesis); Swarm/meta hub expanded to all 14 entries
v0.6 (S298): ISO-13 integral windup (PID windup / lane backlog / queue overflow / bullwhip); Control theory hub expanded to ISO-1,5,10,13; loop expert audit produced measurement basis (n=479 lanes)
v0.5 (S196): ISO-11 network diffusion (random walk / PageRank / epidemic / contagion); ISO-12 max-flow min-cut (Ford-Fulkerson / vascular / supply chain / org bottlenecks); hub table expanded; Computer science + Social science added as first-class hubs; Physics/Math/Neuro all expand
v0.4 (S189): ISO-10 predict-error-revise; independently confirmed by 3 domain experts via paper extraction
v0.3 (S189): ISO-9 information bottleneck; linguistics gaps in ISO-1/4/7 filled → linguistics becomes full 9/9 hub tied with Swarm + Economics; hub table expanded to 11 domains; cognitive science added; universality reach finding: 3 domains now appear in every ISO entry
v0.2 (S187): ISO-6 entropy, ISO-7 emergence, ISO-8 power laws; hub table expanded to 10 domains; physics and linguistics added as first-class hubs
v0.1 (S187): 5 seed entries, 7 hub domain candidates, 6 open questions