0005 — The Institutional Standards: the third doctrinal layer

Date: 2026-07-31 Status: accepted Extends: 0001 (Codices as supreme authority), 0003 (the Registry), 0004 (Codex 0 as a specification book).

Context

After 0004, Anabasis had two doctrinal layers. Codex 1 says what must endure. Codex 0 and Codices 2–10 say how the institution works, in citable clauses, with a Registry that gates what may be built.

Nothing said how things must be built, or how a reviewer proves a thing was built to standard rather than merely argued for well. The repository had no definition of "done", no conformance instrument outside the Codex 6 UX rubric, and no consistent way to name a deviation. Doctrine that cannot be checked drifts, and an institution intended to last generations cannot rely on the memory of whoever happens to be present.

Large, long-lived organisations avoid fragmentation with three layers: a constitution, an explanation of how the organisation works, and a building code. The third layer was missing.

Decision

1. A Standards layer is established at docs/standards/, indexed by 00-index.md, sitting below the Codices and above decision records and implementation. Eight standards:

2. The defining rule of the layer: every standard clause is checkable by a reviewer who did not do the work. A clause whose satisfaction cannot be verified by a second person is not a standard; it is principle, and belongs in the governing Codex. This is what distinguishes a Standard from a Codex: a Codex says what is true and why, a Standard says what you must do and how it is verified.

3. Fixed normative vocabulary — must, must not, should, may — defined once in the index and binding across all eight.

4. Waivers are written or they do not exist. A must clause is waived only with what was traded, what repays it, an owner, and an expiry. A must not clause is never waived. A pattern of waivers against one clause is evidence the clause is wrong, and it is amended rather than quietly ignored.

5. STD-Q is the single gate. Nothing reaches a member, client, partner, the public, or production data without passing it. It references the other seven rather than restating them, and it ends in a conformance certificate a reviewer completes.

6. The Codex 6 rubric is not duplicated. STD-X points at docs/codices/CODEX-6/RUBRIC.md and its templates as its scoring instrument.

Reconciliation performed alongside

1. 1Chapter 04 gains §04.4.20 Board member and §04.4.21 Guest with their visibility contracts, and two open questions: board decision rights (Chapter 03 currently assigns all decision rights to the two Offices) and guest expiry defaults.
2. 2The Registry reserves AGT-06 through AGT-14 — Business, Financial, Writing, Career, Founder, Executive, Knowledge, Curriculum Builder, Risk — as IDs only, each requiring a full Chapter 13 specification before implementation. A separate administrative agent was deliberately not reserved, to keep the number of systems with operational data access small.
3. 3Chapter 07 gains §07.2.0, the four top-level categories — People, Organizations, Knowledge, Capital — as the data model's organizing frame, with Event and AuditRecord named as cross-category records about the four rather than members of one.

Consequences

Every future change now carries a conformance step. That is a real, permanent cost, paid on every piece of work, and it is the intended trade: it is cheaper than rebuilding what was never specified, and it is what allows a contributor in twenty years to produce work indistinguishable in quality from today's.

The risk accepted is bureaucracy — standards obeyed in form and not in substance. The mitigation is §2.2 of the index: unverifiable clauses are removed. A checklist of unverifiable lines is theatre, and the layer is designed to reject it.

Rejected alternative: folding these rules into Codices 2–10. This is where they nominally lived, and it is why they were unenforceable — they were written as principle, mixed with explanation, with no conformance instrument and no waiver procedure.

Institutional Critic (Codex 10)

1. 1Failure at 100x. Eight standards maintained by many hands drift apart and contradict each other. Mitigation: STD-Q is the only gate and references the rest, so contradiction surfaces at the point of use rather than in a document nobody reads; clause numbering is permanent; every amendment is a decision record.
2. 2Unverified assumptions. That reviewers will be independent of authors. Unverified while the institution is small and the same person often does both. Named in STD-Q as the review independence rule, with a recorded exception rather than a pretence.
3. 3Simpler solution. One combined standards document. Rejected: subject standards are consulted by different people at different moments, and a single file would be read by nobody. The index gives the single entry point instead.
4. 4Duplication. Real risk against Codices 5, 6, 7, 8. Handled by the cite-don't-restate rule and by pointing STD-X at the existing rubric rather than reproducing it.
5. 5Enterprise view. Strongly favourable. A documented, waivable, auditable building code is what security review, procurement, and partner diligence ask for, and it exists before it is demanded.
6. 6First-time contributor. Adds reading before the first commit. Mitigated: read the index, then one subject standard, then STD-Q. Three documents, not eleven.
7. 7Legal, operational, security. These are internal engineering documents. The index §8 forbids quoting them as compliance, certification, or audit claims. STD-S materially improves the security posture by making the enforcement obligations checkable.
8. 8Debt. The eight standards were authored in parallel and need a cross-reference consistency pass; each carries its own open questions rather than concealing them. Repaid by the consistency pass recorded in this decision's implementation.
9. 9Consistency. Nothing here alters Codex 1. §1.2 of the index makes Codex subordination explicit and forbids resolving a Codex-Standard conflict in code.

Codex 8 evaluation

Revenue potential: indirect — enterprise and partner diligence, and the avoided cost of rework. Implementation cost: authoring time; no runtime cost. Maintenance cost: real and permanent, and paid on every change. Operational complexity: adds one required step to every piece of work. Customer value: indirect, through consistency of what they receive. Enterprise value: high. Long-term strategic value: high — this is the layer that keeps the institution recognisably itself as the number of hands grows.