The Last Application · Part V — Technical Appendices

Chapter 19 — Graph Dynamics and Integrity

3 min read

Chapter 18 fixed GG as the model. This chapter defines change: proposals, integrity, commit, events, and projections.

Mutations

Definition 19.1 (Mutation). A mutation Δ\Delta is a finite triple

Δ=(C,U,D)\Delta = (C, U, D)

where CC is a set of objects to create, UU a set of (id,d)(\mathrm{id}, d') payload updates (and optional link rewrites), and DD a set of identifiers to delete.

Write Δ|\Delta| for the number of affected objects. The empty mutation is Δ=(,,)\Delta = (\emptyset, \emptyset, \emptyset).

Definition 19.2 (Application). Given well-formed GG and Δ\Delta, the raw application GΔG \oplus \Delta is the graph obtained by inserting CC, applying UU, and removing DD (and all edges incident to deleted nodes). Raw application need not preserve Axioms 18.1–18.3.

Integrity

Definition 19.3 (Integrity predicate). An integrity predicate is a boolean

I(G,Δ){0,1}.I(G, \Delta) \in \{0, 1\}.

Δ\Delta is admissible at GG when I(G,Δ)=1I(G, \Delta) = 1.

Definition 19.4 (Four integrity classes). Decompose

I(G,Δ)  =  IstrItypeIrefIsemI(G, \Delta) \;=\; I_{\mathrm{str}} \wedge I_{\mathrm{type}} \wedge I_{\mathrm{ref}} \wedge I_{\mathrm{sem}}

where, writing G=GΔG' = G \oplus \Delta:

  1. Structural IstrI_{\mathrm{str}}GG' has no forbidden cycles or orphan components relative to the structural rules of Σ\Sigma (for example: every page reachable from an experience root; no component parenting cycles).
  2. Typing ItypeI_{\mathrm{type}} — for every oV(G)o \in V(G'), d(o)L(o)d(o) \models L(o).
  3. Referential IrefI_{\mathrm{ref}} — every edge target in GG' exists in V(G)V(G').
  4. Semantic IsemI_{\mathrm{sem}} — all declared domain invariants and policy constraints hold (quality gates, brand rules, environment policy, constitutional flags).

Ideal. Commit only when I=1I = 1.

Runtime approximation. A deployed system may evaluate III^\circ \le I (for example typing plus soft referential checks) and treat some failures as warnings rather than hard rejects. The mathematics of the paradigm remains the full II; II^\circ is an implementation gap to close, not a different theory (same claim in Chapter 21).

The transition δ\delta

Definition 19.5 (Gated transition). The commit map is

δ(G,Δ)  =  {GΔif I(G,Δ)=1Gotherwise.\delta(G, \Delta) \;=\; \begin{cases} G \oplus \Delta & \text{if } I(G, \Delta) = 1 \\ G & \text{otherwise.} \end{cases}

Definition 19.6 (Action pipeline). An action aa (craft step, workflow step, API mutation, heal) induces a proposal Prop(a,G)=Δa\mathrm{Prop}(a, G) = \Delta_a. The full pipeline is

G    a    Δa    I    δ(G,Δa)  =  G.G \;\xrightarrow{\;a\;}\; \Delta_a \;\xrightarrow{\;I\;}\; \delta(G, \Delta_a) \;=\; G'.

Mnemonic: execute → propose → validate → commit.

Proposition 19.1 (Closure under gated transition). If GG satisfies Axioms 18.1–18.3 and II includes at least ItypeIrefIstrI_{\mathrm{type}} \wedge I_{\mathrm{ref}} \wedge I_{\mathrm{str}}, then G=δ(G,Δ)G' = \delta(G, \Delta) satisfies the same axioms whenever I(G,Δ)=1I(G, \Delta) = 1.

Proof sketch. Typing and referential classes are exactly Axioms 18.2–18.3 on GG'; structural class restores well-formed reachability; identity injectivity is preserved if creates mint fresh ids and updates do not rebind id\mathrm{id}. \square

Events

Definition 19.7 (Event). An event ee is an object with schema in the event family, recording a transition:

e=(id,t,actor,kind,payload)e = (\mathrm{id},\, t,\, \mathsf{actor},\, \mathsf{kind},\, \mathsf{payload})

with timestamp tt and a payload sufficient to audit Δ\Delta (or a redacted digest).

Ideal. Every successful commit emits at least one event. Incomplete emit paths are holes in organizational memory — R(S)R(S) that cannot see its own past.

Definition 19.8 (Issue). An issue is a judgment object opened by a workflow or gate, distinct from a raw event: events say what happened; issues say what must be resolved.

Projections

Definition 19.9 (Projection). A projection is a map

π:GraphView\pi: \mathbf{Graph} \to \mathsf{View}

computable from GG (and configuration in XX) alone. Examples: Experience HTML, GraphQL responses, IDE page trees, Assistant retrieval contexts, brand CSS variables.

Axiom 19.1 (Disposable views). If π(G)\pi(G) is lost, it is regenerable from GG. No projection is a second source of truth. Editing π(G)\pi(G) without updating GG is exactly FF_\perp.

Operators and admissibility

Definition 19.10 (Operator). An operator O\mathcal{O} is a (possibly nondeterministic) map that, given GG and a context cc (user, run, environment), produces a proposal Δ=O(G,c)\Delta = \mathcal{O}(G, c).

Governance (Chapter 23) restricts operators to an admissible set A\mathcal{A}. Semantic closure under governance is

OA:I(G,O(G,c))=1    δ(G,O(G,c)) remains descriptively complete.\forall\, \mathcal{O} \in \mathcal{A}:\quad I\big(G,\, \mathcal{O}(G, c)\big) = 1 \;\Rightarrow\; \delta\big(G,\, \mathcal{O}(G, c)\big) \text{ remains descriptively complete.}

The point

GG moves only by gated δ\delta. Integrity is four predicates. Projections read; they do not author. The next chapter specializes operators to workflows — BPM-class process and agentic orchestration — as mathematical objects on the same graph.