Morphism Philosophy

Source: morphism-philosophy.md (ingested 2026-03-28)

type: normative authority: canonical audience: [agents, contributors, founders] last-verified: 2026-03-26 scope: governance style: C

Morphism Philosophy

Status: Accepted Authority: This document is normative. It defines the axioms from which the Kernel invariants are derived. Derived artifact: morphism-kernel.md Rationale: ADR-002 · ADR-004

This document explains WHY the invariants exist. It does not restate them. Normative invariant definitions are in morphism-kernel.md.

Formal counterpart: The 5 philosophy axioms here motivate the 10 formal axioms in Paper 1 (Agentic Mathematics, papers/agentic-mathematics/main.tex, Section 3). See papers/axiom-resolution.md for the mapping between philosophy and formal axioms.

Axioms

Five axioms. Each is a worldview claim, not an operational rule. The Kernel translates these into enforceable invariants. The tenets translate invariants into daily practice.

A-1 · Entropy Is the Default

Every system decays toward incoherence without active governance. Disorder increases unless work is applied to maintain order. In software this is drift. In LLM sessions it is context loss. In documentation it is duplication that silently diverges.

Governance does not prevent entropy. It makes entropy measurable and its increase visible.

Generates: I-2 (Drift Is Debt), I-5 (Entropy Monotonicity)

A-2 · Structure Over Process

Governance expressed as mathematical structure is composable and verifiable. Governance expressed as process checklists is fragile and non-portable.

A policy encoded as a functor composition works identically across CLI, MCP, CI, and Dashboard. A policy encoded as a checklist works only where someone remembers to run it.

Generates: I-1 (One Truth Per Domain), I-3 (Observability)

A-3 · Evidence Over Trust

Compliance is a computed property, not a social agreement. A proof is a typed artifact, not a log entry. Trust is an output of evidence, not a precondition.

When an agent claims a check passed, the system requires the proof object. When a human claims a doc is current, the system requires the hash match. Claims without evidence are undefined.

Generates: I-3 (Observability), I-6 (Refusal as Structure)

A-4 · Refusal Is Work

Rejecting invalid work is as valuable as accepting valid work. Systems that make rejection expensive or invisible produce silent failures.

A governance system that can only say "yes" or "error" is incomplete. Structured refusal — with violated tenets, required inputs, and alternative paths — is a first-class output.

Generates: I-6 (Refusal as Structure)

A-5 · Scope Is a Boundary, Not a Suggestion

Implicit authority expansion is the primary mechanism by which governance erodes. Every protocol must declare what it does NOT govern. Undeclared scope is not inherited — it is rejected.

The cost of refusing out-of-scope work is low. The cost of accidentally governing a domain without authority is drift that compounds silently.

Generates: I-4 (Scope Binding), I-7 (Minimal Authority)

Why Category Theory

The governance engine uses category theory — morphisms, functors, natural transformations — as its formal foundation. This is not decoration. It is what makes three properties possible:

Composability. A policy expressed as a functor composition works identically across surfaces. Flat rule lists do not compose. Functor chains do.

Completeness detection. The category structure identifies gaps: a missing morphism between policy and enforcement is a provable absence, not an unnoticed omission. Process checklists cannot detect their own gaps.

Migration safety. Natural transformations allow policy migration between versions. Without this structure, schema changes break enforcement silently. With it, the engine can prove whether a change is compatible or breaking.

Mathematical exposition: Paper 1 (papers/agentic-mathematics/main.tex)

The Minimal Generating Set

The Kernel defines exactly seven invariants. Not six. Not eight.

Every axiom generates at least one invariant. Every invariant traces to at least one axiom. No axiom is redundant — removing one leaves at least one invariant without motivation.

AXIOM                          INVARIANT
─────                          ─────────
A-1  Entropy Is the Default    I-2  Drift Is Debt
                               I-5  Entropy Monotonicity

A-2  Structure Over Process    I-1  One Truth Per Domain
                               I-3  Observability

A-3  Evidence Over Trust       I-3  Observability
                               I-6  Refusal as Structure

A-4  Refusal Is Work           I-6  Refusal as Structure

A-5  Scope Is a Boundary       I-4  Scope Binding
                               I-7  Minimal Authority

I-3 and I-6 each have two axiom sources. This is intentional — observability is motivated by both the need for structural evidence (A-2) and the requirement that evidence replaces trust (A-3). Refusal is motivated by both the value of rejection (A-4) and the requirement that rejection produce evidence (A-3).

The derivation from invariants to the ten operational tenets is proven in Paper 1 Section 10 (papers/agentic-mathematics/main.tex) and fully traced in papers/bible/full-proofs.md.

Scope

This document governs: the axiomatic worldview from which the Kernel invariants are derived; the justification for category theory as the governance formalism; the minimal generating set argument.

This document does NOT govern: invariant definitions (Kernel), operational tenets (Doctrine Section 12), enforcement scripts (Doctrine Section 10), mathematical formalism (Paper 1), product positioning (about/).

Amendment Process

Open a PR modifying this file and the corresponding ADR.
Obtain review from CODEOWNERS.
Merge only after drift-check and integration-gate pass.

Amendments that change an axiom MUST include an impact assessment on every Kernel invariant the axiom generates. If the invariant set must change as a result, the Kernel amendment process applies.

Derived artifacts: morphism-kernel.md Rationale: ADR-002 · ADR-004