0004 — Codex 0 becomes a specification book; Codex 6 gains a UX rubric

Date: 2026-07-31 Status: accepted Supersedes: nothing. Extends 0001 (Codices as supreme authority), 0002 (Institutional UX, financial evaluation, the Critic), 0003 (the Registry).

Context

Codex 0 existed as a single-file institutional map plus the Registry. The map answered what exists well enough to gate new work, but it did not specify systems to the depth an engineer, designer, advisor, or AI agent could build from without re-deciding architecture each time. Doctrine that must be re-derived is doctrine that will drift.

Separately, Codex 6 stated the experience doctrine — confidence is the product, one primary action per view, progressive disclosure, cognitive load is a defect, mastery over gamification — as prose. Prose is arguable. Two reviewers could both believe they were applying it and ship inconsistent surfaces.

Decision

1. Codex 0 becomes a multi-chapter specification book at docs/codices/CODEX-0/, twenty numbered chapters: index, institution, divisions, offices, people, CapabilityOS, knowledge graph, data model, permissions, workflows, API boundaries, surfaces, dashboards, AI agents, tooling, finance, reporting, engineering, scale and expansion, glossary. Each chapter carries permanent clause numbers (§07.4.2) so doctrine can be cited stably for decades. Every chapter states its interfaces, invariants, prohibitions, and open questions, so what is undecided is visible rather than assumed.

The Registry (CODEX-0-REGISTRY.md) remains the gate; the chapters are what the gate's eight questions are answered from. The former single-file map is retained for orientation and is subordinate to the chapters.

2. Codex 6 gains an enforceable rubric at docs/codices/CODEX-6/: an eight-dimension 0–3 score with written evidence, plus a surface template, an interaction template, and a reviewer's verdict form. Any dimension at 0 blocks; a 1 ships only with a named owner and date; 16 or below is redesigned rather than patched. The final gate is unchanged and unscoreable: did this make the person's life better?

Consequences

Accepted cost: doctrine is now large enough that nobody reads it end to end. Mitigated by the index, permanent clause citation, and per-chapter scope statements — the intended use is targeted reading, not sequential reading. A second cost is that every new surface now carries specification and review overhead before implementation. That is the point; it is cheaper than rebuilding a surface that was never specified.

Rejected alternative: keeping doctrine as principles and trusting judgement at implementation time. This is what produces institutions whose early years are coherent and whose later years are not.

Institutional Critic (Codex 10)

1. 1Failure at 100x. The chapters become internally inconsistent as they are amended independently. Mitigation: clause numbers never move, cross-references are by clause, and Codex 9 requires a decision record for every amendment.
2. 2Unverified assumptions. That contributors will read chapters rather than guess. Unverified until there are contributors; the AGENTS.md binding covers the AI case now.
3. 3Simpler solution. A shorter Codex 0. Rejected: brevity is what forced re-derivation.
4. 4Duplication. Chapters restate material from Codices 2–8 by design; the domain Codices remain the doctrine, the chapters the specification. Where they conflict, the domain Codex governs on principle and the chapter on structure.
5. 5Enterprise view. Favourable: a documented data model, permission model, and reporting scope are what procurement and security review ask for.
6. 6First-time contributor. Twenty chapters is intimidating. Mitigated by the index and by the rubric templates, which tell a newcomer exactly what to produce.
7. 7Legal, operational, security. The chapters are internal doctrine, not compliance or certification claims; README §Status says so. The permissions and data chapters strengthen the security posture by making "never sees" explicit and server-side.
8. 8Debt. Chapters 06 through 16 were authored in parallel and need a consistency pass on cross-references and entity naming. Repaid by that pass; open questions are recorded in each chapter rather than hidden.
9. 9Consistency. Nothing here alters Codex 1. The rubric encodes Codex 1 Article VI's prohibitions as interface rules.

Codex 8 evaluation

Revenue potential: indirect — enterprise and partner diligence. Implementation cost: authoring time only, no runtime cost. Maintenance cost: real and permanent; every architectural change now updates a chapter. Operational complexity: low. Customer value: indirect, through consistency of what they receive. Enterprise value: high. Long-term strategic value: the highest available — this is the document set that lets the institution be recognisably itself after its founders.