The Last Application · Part V — Technical Appendices

Chapter 23 — Control Plane and Operational Rules

4 min read

Mathematics without admissibility is a rewrite engine. This chapter defines environments, roles, admissible operators, the promotion ladder, and the IDE/Console split — the institutional form of “under governance, while running.”

Environments as indexed graphs

Definition 23.1 (Environment). Let $\mathsf{Env}$ be a finite set of environment labels. Typical:

\mathsf{Env} = \{\mathsf{Personal},\, \mathsf{Dev},\, \mathsf{Test},\, \mathsf{Prod}\}.

For each $e \in \mathsf{Env}$ , write $G_e$ for the product graph (or pinned view) visible in that environment. Personal graphs are per principal; Dev/Test/Prod are organization-scoped.

Definition 23.2 (Pin / release). A release is an immutable snapshot $R^\star = \mathsf{Snap}(G_{\mathsf{Dev}})$ . A pin binds an environment to a release:

\mathsf{pin}: \{\mathsf{Test},\, \mathsf{Prod}\} \to \mathsf{Releases}.

Admissible operators

Definition 23.3 (Context). An operator context is

c = (\mathsf{actor},\, e,\, \mathsf{role},\, \mathsf{caps})

with capabilities including an MCP/IDE flag.

Definition 23.4 (Admissible set).

\mathcal{A}(c) \subseteq \mathsf{Operators}

is the set of operators actor may invoke on $G_e$ . In the worked example, $\mathcal{A}(c)$ is computed from active schema:policy documents (deny $>$ require_human $>$ allow) over Principal × Action × Resource × Condition. Bootstrap Policies encode the historical role×pillar matrix; orgs may refine further. Viewers typically have $\mathcal{A}$ empty for write operators.

Axiom 23.1 (No ungated prod craft). Direct write operators on $G_{\mathsf{Prod}}$ lie outside $\mathcal{A}$ for ordinary builders. Production changes enter through promote (below).

Ladder

Definition 23.5 (Merge). A merge request is a proposal

\mathsf{MR}: G_{\mathsf{Personal}} \rightsquigarrow G_{\mathsf{Dev}}

subject to Dev policy (including $N$ -approver thresholds and change windows).

Definition 23.6 (Promote). A promote request is a proposal to update a pin:

\mathsf{PR}: R^\star \mapsto \mathsf{pin}(e), \quad e \in \{\mathsf{Test},\, \mathsf{Prod}\}

subject to policy, windows, and quality gates (Issue severity, craft coverage, …) — instances of $I_{\mathrm{sem}}$ .

Personal --merge--> Dev --release--> snapshot
                         |
                    promote (HITL / window / N-approvers)
                         v
                   Test pin --> Prod pin

Definition 23.7 (Change window). A window $W_t \subset \mathsf{Time}$ restricts when merge/promote operators are in $\mathcal{A}$ . Outside $W_t$ , those operators are undefined (rejected).

IDE builds; Console governs

Definition 23.8 (Surface split).

Surface	Allowed effects
IDE (tools over MCP)	Operators in $\mathcal{A}$ on Personal/Dev: craft, scaffold, build, verify, submit merge/promote requests, fulfill $\mathsf{waiting\_ide}$
Console	Approvals, policy, team, brand settings, Trust center, audit — govern $\mathcal{A}$ and queues

Axiom 23.2 (Secrets). Secret collection is an operator that returns only sealed handles. Plaintext secrets are never arguments to IDE tools and never fields in ungated chat.

Tool surface as typed operators

Definition 23.9 (IDE tool). A tool is a typed signature

\mathsf{tool}: \mathsf{Args} \to \mathsf{Result}

that, when authorized, induces $\mathcal{O} \in \mathcal{A}(c)$ and thus a proposal $\Delta$ . Examples of tool classes:

Inspect: $\pi_{\mathrm{IDE}}$ reads (status, experience tree, taxonomy, governance status)
Mutate: craft, unlink, delete, scaffold
Verify: run deterministic $V$ (Chapter 20 loops); optional remediate
Build: start workflow runs; agent task fetch/submit for $\mathsf{waiting\_ide}$
Collect: open sealed form runs
Govern: submit merge; read queues (approve remains Console)

Trust center

Definition 23.10 (Governance hub). The Trust center is the Console projection of:

\mathsf{Gov} = (\mathsf{Identity},\, \mathsf{Change},\, \mathsf{Assurance},\, \mathsf{Integrations},\, \mathsf{Agents},\, \mathsf{Evidence})

posture over $\mathcal{A}$ , ladders, gates, sealed connectors, agent wait states, and audit export. It is not a promote-only queue; it is the human-readable face of admissibility.

Skills as Knowledge

Definition 23.11 (Skill pull). Let $\Sigma_{\mathrm{skill}} \subset G$ . An IDE session with stale marker $v$ refreshes when

v < \mathsf{version}(\Sigma_{\mathrm{skill}})

and materializes doctrine locally as a cache of $\pi_{\mathrm{IDE}}(\Sigma_{\mathrm{skill}})$ . The SoT remains $G$ .

Closing fixed point

The paradigm is a fixed point of three maps:

\begin{aligned} \mathsf{Math}&: S \mapsto (G,\, \delta,\, I,\, W,\, \rho) \\ \mathsf{Arch}&: G \mapsto \{\text{pillars},\, \pi_j,\, \mathsf{Store}\} \\ \mathsf{Ctrl}&: (G_e,\, c) \mapsto \mathcal{A}(c) \end{aligned}

A system is semantically closed in the operational sense when every product change is an operator in some $\mathcal{A}(c)$ , every such operator’s proposal passes $I$ before commit, and every facet of the product lies in $G$ .

Operational rules (summary)

Graph is SoT — projections and local files are not.
IDE builds; Console governs.
Secrets only via sealed collect.
Verify before ship — deterministic $V$ ; loops budgeted; HITL on exhaust.
One org brand stack — Experiences inherit design system.
Events and Issues are memory and judgment on (or keyed to) $G$ .
Harden $I$ toward enforce on the product path — that is the work, not a different paradigm.

The point

Chapter 3 stated seven clauses in English. Through Chapter 23, Part V states them as $F \in R(S)$ , a typed graph $G$ , gated $\delta$ , process and agentic workflows $W$ , pillars and projections $\pi$ , and admissible operators $\mathcal{A}$ . Chapters 24–25 give the same objects in JSON-LD and walk a concrete run. When the runtime is thinner than $I$ , the mathematics still names the gap — and the product path is to close it.