Coherence Theory

A complete framework for understanding why patterns persist. From atoms to organizations, one law governs survival.

1. Primitive Ontology

Coherence Theory starts from five primitive concepts. Everything else — quantum mechanics, general relativity, Darwinian evolution — is derived from these foundations.

Patterns

Plain: Any regularity you can recognize more than once.

Formal: Re-identifiable regularities at finite resolution. A product is a pattern. A user's workflow is a pattern. A market gap is a pattern. Code is a pattern. You are a pattern.

Pokes

Plain: Disturbances from the environment. Things that happen to a pattern.

Formal: Local disturbances from neighboring patterns. User complaints are pokes. Competitor launches are pokes. API failures are pokes. Every poke has bounded reach — no instantaneous action at a distance.

Ticks

Plain: A repeatable reference event used to measure time. There is no universal clock.

Formal: Repeatable reference pokes. A deploy cycle is a tick. A sprint is a tick. Time is always measured relative to some pattern's regular behavior — there is no global clock, only mutual tick-counting between patterns.

Coherence (CL)

Plain: How well a pattern holds together under stress. High coherence means robust. Low coherence means fragile.

Formal: The degree to which a pattern preserves its defining regularities under worst-case pokes. A product with high CL solves its problem reliably across edge cases, user types, and environmental changes.

Budgets

Plain: The costs of staying alive. Every pattern pays to persist, and the costs come in exactly three independent dimensions.

Formal: Three independent, orthogonal costs derived from discrete Hodge decomposition on the contact graph. Why exactly three? Fewer cannot stabilize open systems. More would contradict finiteness. This is a mathematical theorem, not an assumption.

Budget	Measures	Minimized By
B_th Throughput	Net routing, transport, I/O, API calls, data movement	Caching, parallelism, canceling redundant work
B_cx Complexity	Internal coordination, abstractions, dependencies, branching	Cycle-free wiring, inlining, deleting code
B_leak Leakage	Boundary exposure, unhandled errors, trust loss	Pointer alignment, insulation, error handling

Critical: These three budgets are orthogonal. B_leak is NOT “wasted B_th.” A pattern can have high throughput with zero leakage (sealed internal transport) or zero throughput with high leakage (fragile boundary, no transport). You cannot fix leakage by improving throughput. They are mathematically independent dimensions.

2. The Selection Inequality

The central theorem of Coherence Theory. For any pattern A with coherence CL(A) and budget vector B(A):

The Selection Functional

A pattern persists if and only if Sel(A) ≥ 0. Patterns on the boundary (Sel = 0) define the coherence frontier.

Think of CL as revenue and the three B terms as three independent cost categories. A business survives when revenue exceeds costs. But the costs are not fungible — you cannot fix churn (B_leak) by cutting operations (B_th). They are independent dimensions.

The Multipliers

The multipliers are prices. They encode how expensive each budget dimension is in the current environment. Only ratios matter — absolute scales wash out (calibration invariance).

The Selected Equalization Point (SEP)

The SEP is the unique point on the coherence frontier where marginal gains per unit budget are equalized across all active dimensions. At SEP, no cost-neutral reallocation of budgets can increase coherence. This is the optimally efficient configuration.

Exchange Equalization at SEP

This is the trade ratio. If you are going to spend one more unit of budget, which dimension gives the most coherence per unit cost? At SEP, the answer is the same for all dimensions.

3. The Ten Priors

These are the metaphysical foundations. Everything in CT is derived from these ten statements. If a conclusion violates a prior, the conclusion is wrong.

Patterns are fundamental. Reality is made of regularities, not stuff.

Formal: There exist re-identifiable regularities P at finite resolution.

Example: The letter "A" is a pattern — you recognize it in any font, size, or handwriting. So is a hydrogen atom, a business model, or a musical phrase.

Falsifier: No regularity can be re-identified across any two observations.

Some patterns persist. Not everything is noise.

Formal: There exists at least one pattern A with CL(A) > 0.

Example: Atoms persist for billions of years. Species persist for millions. Companies persist for decades. Something is holding them together.

Falsifier: All CL = 0: pure noise everywhere, no stable structures at any scale.

Existence is relational. A pattern is defined by its neighborhood, not in isolation.

Formal: CL is defined relative to neighborhood N(A), not absolutely.

Example: A startup's coherence depends on its market, competitors, and customers. Move the same team to a different market and their coherence changes.

Falsifier: A perfectly isolated pattern has measurable coherence.

Disturbances are local. Nothing acts everywhere at once.

Formal: Bounded-support poke cone; ticks are pokes with finite reach.

Example: A server crash affects your app, not every app in the world. A competitor launch affects your market segment, not all markets. Influence propagates at finite speed.

Falsifier: Instantaneous nonlocal influence — a change here instantly affects something arbitrarily far away.

Selection pressure exists. Some patterns survive better than others.

Formal: Survival frequencies differ; coherence discriminates among patterns.

Example: Not all startups survive. Not all species survive. Not all codebases survive. Coherence is what separates survivors from failures.

Falsifier: All patterns survive equally regardless of their properties.

Persistence has a cost. Nothing is free.

Formal: There exist budgets B >= 0 constraining persistence.

Example: Running a server costs money. Maintaining a species costs metabolic energy. Keeping a team aligned costs communication overhead. Every pattern pays to exist.

Falsifier: A pattern maintains its identity without any constraint or cost.

Budgets are finite. Nothing can endure unlimited stress.

Formal: There is a finite bound on admissible stress before CL drops.

Example: Every bridge has a load limit. Every server has a request limit. Every person has a stress limit. Push past it and the pattern breaks.

Falsifier: A pattern endures unlimited pokes without degradation.

Costs are multidimensional. You cannot collapse all costs into one number.

Formal: Budget space has dimension >= 2 (= 3 by classification theorem).

Example: A company has operating costs (B_th), organizational complexity (B_cx), and customer churn (B_leak). These are different problems requiring different solutions.

Falsifier: A single scalar captures all persistence costs for all patterns.

No pattern is complete. There is always an uncaptured disturbance direction.

Formal: There exist disturbance directions not captured by current essentials.

Example: Every software system has undiscovered bugs. Every business plan has assumptions that will be wrong. Every scientific theory has domains it has not been tested in.

Falsifier: Find a pattern that endures all possible disturbances indefinitely — a perfectly complete system.

A10

Adaptation is required. Static patterns die.

Formal: Survivors reorganize to maintain Sel >= 0 under changing conditions.

Example: Companies that do not adapt to market changes fail. Species that do not adapt to environmental changes go extinct. The only constant is that you must keep changing.

Falsifier: A static, unchanging pattern persists indefinitely in a changing environment.

4. The Seven Operational Axioms

Derived from the priors, not independently assumed. These are the working tools for applying CT to real systems.

Axiom	Name	Meaning for Products
B1	Convexity, LSC, Coercivity	Budget functionals are well-behaved; optimization has solutions. You can always find the best configuration.
B2	Functoriality	Processing can only destroy coherence: B(f of A) <= B(A). Every transformation leaks. Data processing never adds information.
B3	Ampliation Invariance	Adding unused features does not reduce costs: B(A tensor I) = B(A). Features must earn their place.
B4	Local Additivity	Independent components' budgets add: B(A1 parallel A2) = B(A1) + B(A2). Decompose to optimize.
B5	Gauge/Ad-Invariance	Cost-neutral relabelings do not change budgets. Renaming variables does not change program complexity.
B6	Quadratic Tangent Law	Near equilibrium, cost is quadratic in deviation. Small changes cost little; large changes cost quadratically more.
B7-R	Uniform Calibration	Only ratios matter; absolute scales wash out. What matters is the relative price of budgets, not their absolute values.

B3 (Ampliation Invariance) is your anti-feature-creep axiom. Adding a feature nobody uses does not reduce any budget. It increases B_cx (more code to maintain) without increasing CL. Every feature must earn its place.

B4 (Local Additivity) is why you decompose products into independent components. If tool A and tool B have disjoint supports, their budgets add. You can analyze and optimize them independently.

5. Domain Organism Theory

Every sufficiently large coherent domain is an “organism” with six necessary structural elements. These are theorems, not metaphors. When analyzing any system, you must identify all six.

Scaffold

The stable ground the pattern operates on. If the scaffold is unstable, nothing else matters.

Formal: A stable pseudo-metric d_eff on the domain. Every hop costs B_th >= epsilon_0 > 0.

In software: The technology stack, deployment infrastructure, hosting platform. If your infrastructure is changing under you, B_th is unbounded per tick.

In business: The market you operate in, the regulatory environment, the platform you build on. A stable scaffold means you can plan; an unstable scaffold means constant firefighting.

Binder

The single most important pattern — the one thing that makes everything else hold together.

Formal: The dominant pattern A* with maximal Sel in the neighborhood. Cascade range is proportional to CL(A*).

In software: The core value proposition. The one feature users come back for. If you cannot state the binder in one sentence, the product has no binder.

In business: The reason customers stay. Not the marketing tagline, but the actual underlying reason. Amazon's binder is convenience. Google's is relevance. What is yours?

III

Loop Networks

Internal feedback cycles that both detect problems and carry information. Loops are simultaneously sensors and transport.

Formal: Internal cycle-space structure with coordination rank beta_1 >= 1. Each loop costs B_cx proportional to beta_1.

In software: CI/CD pipelines, error monitoring, user analytics feeding product decisions. Every feedback loop costs complexity but provides detection and response.

In business: Customer feedback to product to customer. Sales data to strategy to execution. Financial reporting to decisions. More loops mean better sensing but higher coordination cost.

Domain Walls

The boundaries between your system and everything else. Surface tension determines how hard it is to cross.

Formal: Interfaces carrying surface tension tau = lambda_leak * sin^2(Delta theta) + o(Delta theta^2). Tension increases with misalignment angle.

In software: APIs, user interfaces, integration points. A product aligned with users' existing workflow has low surface tension (easy adoption). One requiring behavior change has high tension.

In business: Competitor boundaries, regulatory boundaries, platform boundaries. Every boundary has a crossing cost. Minimize it where you want flow; maximize it where you want protection.

Hidden Editors

Patterns that detect and correct misalignment between sub-domains. Quality control at the seams.

Formal: Chiral controller patterns mediating alignment repair across sub-domain boundaries.

In software: Automated testing, QA, error monitoring, user feedback collection. Any system that detects when things go wrong and triggers correction.

In business: Quality assurance processes, customer success teams, internal auditing. If your editors cannot keep up with misalignment generation rate, sub-domains decohere.

Non-Zero Leakage

There is always loss. Always bugs, always churn, always drift. A system claiming zero leakage is either lying or unobservable.

Formal: B_leak(D) > 0 on any nontrivial lens. This is a theorem from A9 (irreducible openness).

In software: There are always undiscovered bugs. Always edge cases. Always users who leave. The goal is not zero leakage but leakage below the survival threshold: Sel >= 0.

In business: Customer churn is never zero. Employee turnover is never zero. Information loss is never zero. Accept this and budget for it instead of pretending it will not happen.

Organism Coherence10%

Incomplete. The organism needs a scaffold and a binder to stabilize.

Why Quality Control Has Blind Spots

Move your cursor to catch the incoming disturbances. Notice what you miss.

Caught: 0Missed: 0

6. Lens Theory

A lens is a pattern used for observation. It has its own budget profile and is subject to selection. Before analyzing anything, you must specify the lens — what you are looking through, and what it costs to look.

The Three Budget Costs of a Lens

B_th(L): How many ticks to run the analysis
B_cx(L): Internal complexity of the lens
B_leak(L): How much ranking information drifts at scope boundary

Lens Levels

Level	Scope	Predictive Power
0	Excludes all neighbors	Zero — useless
1	Local only, no binder	Fragile — breaks under competition
2	Includes binder	Robust for medium windows
3	All patterns above threshold	Captures competing pressures
4	SEP-calibrated	Maximally coherent

The Binder Inclusion Lemma: A lens that excludes its local Pareto binder fails its own selection criterion. The leakage cost of excluding the binder scales quadratically, while inclusion costs scale linearly. Always include the dominant pattern in your analysis scope.

Quick Self-Test

Pick any system you know well — your company, your codebase, your morning routine. Can you answer these five questions?

1Can you name the binder in one sentence? (The single thing that makes this system persist.)

2Can you identify which of the three budgets is the bottleneck (H_min)?

3Can you name at least one feedback loop?

4Can you identify the domain walls — the boundaries where surface tension is highest?

5Can you name the irreducible leakage? (A9: it is never zero.)

If yes to all five: you understand CT well enough to use it. Proceed to the formal papers for the mathematical backbone, or try applying CT to a real problem.

← Home Papers →