Glossary

Glossary / Lexicon

Shared vocabulary for the SAE Framework. [Contributions Welcome ✍️] — add terms as the playbook grows.

---

Sovereign Architect‑Engineer (SAE) — A single operator who designs, builds, tests, and ships full‑lifecycle software by orchestrating AI agents instead of coordinating a team. The apex role of the Digital Org Chart.

Conductor — The human in the SAE loop. Owns architecture, edge‑case verification, structural‑drift review, and final decisions. See Module 3.

Visualizer — The generative‑UI persona. Turns product intent into semantic component code.

Engineer — The repo‑aware code‑generation persona (e.g. Cursor, Claude Code). Reads specs and the guardrails file (.cursorrules or CLAUDE.md), writes the codebase. See Agent Setup & Configuration.

Guardian — The autonomous‑QA persona whose job is to break what the Engineer builds, in a fresh/independent context. The framework's built‑in check against hallucinated correctness.

The 12 Layers of Bureaucratic Alignment — The legacy handoff chain from executive mandate to change‑management sign‑off that the SAE collapses. See The 12 Layers of Friction.

Brooks's Law — The principle that communication channels grow as $C = \tfrac{N(N-1)}{2}$, making coordination cost scale $O(N^2)$ with headcount. The mathematical root of the friction layer.

Single‑Shot Vertical Slice — Building a feature end‑to‑end (schema → API → state → UI) in one unified generation cycle, so layers stay aligned by construction. See Module 2.

Bounded Context — A fully self‑contained feature/domain that an agent can understand and modify without parsing unrelated parts of the system. Borrowed from Domain‑Driven Design; central to AI hygiene.

AI Hygiene / AI Readability — Structuring a codebase (small cohesive files, strict isolation, self‑documenting layout) so models read and edit it with high precision.

Spec‑Driven Development (SDD) — The practice of treating a Markdown specification — not the code — as the source of truth. Bugs are fixed by editing the spec first, then regenerating. See SDD Spec Template.

The Abstract Manifest — The pristine Markdown spec the operator authors as the system mandate; the input the Engineer generates from.

Zero‑Ops — A tooling principle: if a technology needs a dedicated specialist to deploy, scale, or tune, it's banned. Provisioning and scaling are the platform's job. See Module 4.

Context Entropy Rating (1–13) — A Fibonacci‑flavoured measure of how much context preparation a task needs before generation. Replaces story points. See Module 5.

Self‑Healing Circle — The automated loop where production telemetry feeds an agent that drafts a spec amendment, which the Conductor approves in one click. See Module 5.

Structural Drift — Subtle divergence between layers (or between code and spec) that creeps in over time; a primary thing the Conductor reviews for.

Blast Radius — How much breaks if a given change is wrong. Paired with Context Entropy to keep low‑prep / high‑stakes changes from slipping past review.

Reality Check — The wiki's convention for an honest callout naming the limits of a bold claim. The framework's credibility mechanism. See P7.

Adversarial Independence — The discipline that the context which wrote the code must never be the sole context that judges it. The Guardian runs in a fresh context and is tasked to refute. The structural form of P3; see The Verification Doctrine.

Grey-Box Verification — Verification with partial internal knowledge: the Guardian sees the spec and the interfaces (the intended contract) but judges from a fresh context, designing adversarial cases rather than confirming the author's mental model. Between black-box (no knowledge) and white-box (full knowledge).

Spec-Drift — Divergence between the spec and the code it is supposed to govern. Spec-drift time — how long that divergence persists before reconciliation — is the framework's leading decay indicator (the "canary"). See KPIs & Metrics.

Blast-Radius Gate — A human approval required before an irreversible action (auth, payments, data deletion, schema drops, external sends) is applied. The operational form of P5; see Risks, Security & Governance.

Context Engineering — The practice of establishing exactly the right context (spec + bounded slice + rules, and nothing more) before generation. Paying down Context Entropy is context engineering; it is the real labor when syntax is free.

Evaluative Depth — Enough competence in a domain to judge and refute generated output, as distinct from generative depth (enough to build it from scratch). The floor an operator must clear in every domain a slice touches.

Augmented Coding — Kent Beck's term for keeping traditional engineering values (caring about code, tests, complexity) while using AI as a power tool. SAE sits squarely in this tradition; contrast with "vibe coding." See Where SAE Sits.

Maturity Level (L1–L5) — A position on the Maturity Model ladder, from L1 (ad-hoc AI coding) to L5 (self-healing / sovereign), each defined by observable behaviour and the metric that proves it.