Exploration: Inter-Layer Edges and Graph Constraints

Status: Exploration. Analyzing the edges between the three methodology layers and what the methodology imposes (or doesn't) on graph structure. Question: What are the actual edges between L1, L2, L3? And does the methodology constrain what kinds of graphs it can produce, or is it open to arbitrary structure?

1. The three domains and their potential edges

We have:

• L1 (Domain Analysis): {Pm, Lv, Dp, Ix, Cp, Ps}
• L2 (Graph Construction): {Ed, Ch, Sb, Mp, Cn}
• L3 (Graph Semantics): {In, Ar, Ty, Ab, Pt, Cv}

Potential edges: L1↔L2, L2↔L3, L1↔L3, and any self-edges (L1↔L1, etc.)

2. L1 → L2: Domain analysis feeds graph construction

2.1 What actually happens

When you do graph construction, you take two completed domain analyses and connect them. But more than that: L2 USES L1's tools internally. Bridge primitives on realization edges get the FULL 12-step treatment — they have partial levels, dependencies, pairs, core triads. L1's vocabulary is the language L2 uses to describe edge substance.

So L1 feeds L2 in TWO ways:

1. L1 outputs become L2 inputs — analyzed domains are what edges connect
2. L1 tools are used within L2 — bridge analysis IS domain analysis applied to edge content

2.2 What kind of edge is this?

Let me test against existing types:

Realization? No substrate gap. L2 isn't physically built from L1. No medium change.

Configuration? Not quite — L2 isn't L1 at specific settings. L2 has its own distinct primitives {Ed, Ch, Sb, Mp, Cn} that aren't just L1 primitives at specific levels.

Decomposition? Closer — L2's Substance (Sb) decomposes into L1's vocabulary when you analyze bridge primitives. The bridge primitives' internal structure IS domain analysis structure {Pm, Lv, Dp, Ix, Cp, Ps}. So Sb at high levels (Sb3+) contains L1 structure inside it.

Role identification? Partially — there are structural role correspondences:

• L1's Pm (Primitive) plays the same structural role as L2's bridge primitives within Sb
• L1's Ps (Position) relates to L2's Cn (Constraint) — positions are what constraints constrain
• L1's Dp (Dependency) within a domain parallels L2's Cn (Constraint) across domains

Something else? Maybe this is an EMBEDDING — L1's vocabulary is embedded within L2 as the language for describing substance. L2 doesn't decompose into L1, and L1 doesn't realize L2. Rather, L1 is the INTERNAL LANGUAGE that L2 uses for its most complex component (bridge analysis).

2.3 The embedding relationship

L2's Substance (Sb):
  At Sb0-Sb1: no L1 content — just "there are bridge primitives"
  At Sb2: L1 vocabulary appears — bridge primitives NAMED (Pm-level)
  At Sb3: L1 structure appears — bridge primitives have PARTIAL LEVELS (Lv), DEPENDENCIES (Dp)
  At Sb4: full L1 applied — bridge primitives have PAIRS (Ix), CORE TRIAD (Cp), POSITIONS (Ps)

So L1 appears INSIDE L2's Sb primitive at higher partial levels. This is an embedding: L1's full vocabulary is contained within one of L2's primitives at sufficient analytical depth.

This is actually something we've seen before: substrate primitives becoming AMBIENT at the surface level. Here, L1's vocabulary isn't ambient — it's EMBEDDED in Sb. The difference: ambient means invisible (assumed but not explicit). Embedded means ACTIVE (used explicitly within the containing primitive).

2.4 Characterization

L1 → L2 edge: Embedding. L1's structural vocabulary {Pm, Lv, Dp, Ix, Cp, Ps} is embedded within L2's Substance (Sb) primitive. At low Sb levels, L1 is invisible. At high Sb levels, L1's full vocabulary is active inside Sb. L2 USES L1 as its internal analysis language for edge content.

Additionally, L1's output (analyzed domains as nodes) provides the INPUT that L2 operates on. This is a separate relationship: L1 PRODUCES what L2 CONNECTS.

So there are actually TWO edges from L1 to L2:

1. Production: L1 output → L2 input (analyzed domains become what edges connect)
2. Embedding: L1 vocabulary → L2's Sb (L1's tools analyze bridge content)

3. L2 → L3: Graph construction feeds graph semantics

3.1 What actually happens

L3 operates on the populated graph that L2 builds. L3's Instance (In) is an analyzed domain (L1 output). L3's Arrangement (Ar) is the topology that L2's edges create. L3 discovers patterns in the graph that L1+L2 together built.

3.2 What kind of edge is this?

Production: L2's output (connected graph of domains) is L3's input (what L3 analyzes).

But there's MORE: L2's edge types (Ch) become something L3 analyzes. L3's Arrangement (Ar) is structured BY L2's edges. L3's Type (Ty) classification partly depends on WHERE domains sit in L2's topology (is this domain at the substrate end of a realization edge? at the abstract end of a role-id edge?).

So L2 doesn't just FEED L3 — L2's STRUCTURE determines what L3 can discover. The edge types, bridge structures, and constraints that L2 establishes are the RAW MATERIAL of L3's pattern discovery.

3.3 Characterization

L2 → L3 edge: Production + Structural determination. L2 builds the graph that L3 analyzes. L2's edge types and topology DETERMINE what arrangements, types, and patterns L3 can discover. If L2 only builds realization chains, L3 only sees chain-shaped arrangements. If L2 builds configuration clusters, L3 sees clusters.

This is important: L3's patterns are shaped by L2's edge types. The invariant topology we found (substrate→surface→ecosystem with context and selection) is shaped by the KIND of edges L2 identified (realization chains with bridge primitives). Different L2 edge types would produce different L3 patterns.

4. L3 → L1: Feedback

4.1 What actually happens

L3's abstractions and patterns generate predictions. These predictions feed back to L1 as expectations for new domain analyses. When you analyze a new domain, L3 tells you what to expect (if it's been classified by type).

4.2 What kind of edge is this?

Feedback. Not production (L3 doesn't produce domain analyses). Not realization. Not configuration. It's a feedback loop — the output of the full analytical process modifies the input to new iterations.

This is the same type of edge discovered in the biology analysis: feedback edges where surface activity modifies the context for future substrate activity (niche construction). Here: L3 activity (pattern discovery) modifies the context for future L1 activity (domain analysis).

4.3 Characterization

L3 → L1 edge: Feedback. L3's patterns and predictions modify how L1 approaches new domains. Not deterministic — L1 can ignore L3's predictions. But the predictions set expectations that L1 can confirm or refute.

5. L1 → L3: Direct feed

5.1 What actually happens

An L1 output (analyzed domain) directly becomes an L3 Instance (In), without necessarily going through L2 first. You can analyze three domains with L1 and compare them at L3 without formally building L2 edges between them.

In practice, this happened: the biology analysis compared biology, entity system, and cognition at L3 (finding invariant topology, tight-loose-tight) based on L1 analyses of each, with L2 edge analysis happening in parallel but not as a prerequisite.

5.2 Characterization

L1 → L3 edge: Direct production. L1 outputs can feed L3 directly. L2 ENRICHES the graph (adds edge structure, constraints, mappings) but isn't strictly required for L3 to operate. L3 can compare instances without formal edge characterization.

This means L2 is optional in some analyses — you can go L1 → L3 directly for comparison, then return to L2 for edge characterization. The flow isn't strictly sequential.

6. Self-edges

6.1 L1 → L1: Recursive analysis

A manifestation (Ps) at sufficient complexity can itself be analyzed as a domain. Git is a manifestation in the entity system lattice AND a domain with its own primitives {blob, tree, commit, ref, branch, remote}. This is the RESOLUTION SWITCH described in methodology.md §5.5.

This is a decomposition edge — the manifestation-as-domain decomposes into finer-grained structure.

6.2 L2 → L2: Edge composition

Edges compose through intermediate domains. Biology→Chemistry→Physics is a composed edge. This is internal L2 composition — edges chaining through shared domains.

6.3 L3 → L3: Meta-analysis

L3 can analyze ITSELF — the graph semantics analysis IS a domain that L3 can treat as an instance. This is the meta-circularity we encountered. It produces the substrate-core/ecological-envelope distinction but doesn't create a fourth substrate.

7. The full inter-layer graph

L1 (Domain Analysis)
  │
  ├── Production ──→ L2 (Graph Construction)
  │                    │
  ├── Embedding ──→ L2's Sb (L1 vocabulary used for bridge analysis)
  │                    │
  ├── Direct prod ─────────────────→ L3 (Graph Semantics)
  │                    │                │
  │                    ├── Prod+Det ──→ L3
  │                    │                │
  │   ←── Feedback ────────────────────┘
  │
  └── Decomposition ──→ L1 (recursive — resolution switch)
  
L2 ──→ L2 (composition — edges chain)
L3 ──→ L3 (meta-analysis — self-application)

Edge inventory:

8. New edge types discovered?

The existing edge types: realization, configuration, role-identification, enrichment, decomposition, feedback, selection.

New types encountered here:

8.1 Production

"A produces what B consumes." L1 produces analyzed domains. L2 produces connected graphs. L3 produces patterns. Each layer's output is the next layer's input.

Is this really a new edge type? Or is it just... how processes compose? In software: function A produces output consumed by function B. That's function composition, not a structural relationship between domains.

Assessment: Production might not be a structural edge type. It's a PROCESS relationship — one operation feeds another. The structural relationship between the DOMAINS is what matters, and for L1→L2, the structural relationship is that L2's inputs ARE L1's outputs. This is more like a dependency (L2 depends on L1 to have something to connect) than a structural correspondence.

If we DO formalize it: production is a DIRECTED DEPENDENCY — B requires A's output to operate. Different from structural dependency within a domain (Dp), which is about what primitives presuppose each other.

8.2 Embedding

"A's vocabulary is used inside B's primitives." L1's full vocabulary appears inside L2's Substance primitive at high levels. This is structurally specific — not ambient (invisible), not decomposition (B doesn't break down into A), not realization (no medium change).

Embedding means: one domain's analytical apparatus is CONTAINED WITHIN a specific primitive of another domain. The contained vocabulary is active — it's used to analyze the substance, not just assumed.

Is this really new? Consider: when the entity system uses typed dispatch (X) to evaluate typed entities (E+I+T), the type system's vocabulary is embedded within X's operation. The type system is a sub-domain embedded in the entity system's Execution primitive.

So embedding might already exist in analyzed domains — we just haven't named it. The entity system's relationship to its type sub-system could be characterized as embedding: the type system vocabulary is embedded within E/I/T primitives.

Assessment: Embedding might be a variant of DECOMPOSITION — where instead of the whole domain decomposing, a specific primitive at high levels contains another domain's full vocabulary. It's decomposition at the primitive level rather than the domain level.

8.3 Structural determination

"A's structure determines what patterns B can discover." L2's edge types shape what L3 sees. This is a CONSTRAINT relationship — not realization (no medium change), not configuration (L3 isn't a setting of L2), but L2's choices about what to analyze constrain L3's discovery space.

Assessment: This might be a variant of CONSTRAINT — not cross-lattice positional constraints (Cn), but cross-layer analytical constraints. What L2 analyzes determines what L3 can find.

9. Does the methodology constrain what graphs it can produce?

9.1 What the methodology imposes

Domain-level constraints:

• Domains must have decomposable structure (identifiable primitives)
• Primitives must pass the three tests (minimality, productivity, recurrence)
• Dependencies filter the lattice — not all primitive combinations are coherent
• Partial levels must correspond to qualitatively distinct configurations

Edge-level constraints:

• Edges must connect analyzed domains (at least proto-edge level)
• Edge types have specific substance expectations (realization → bridge primitives, etc.)
• Cross-domain constraints further restrict compatible positions

Graph-level constraints:

• Edges compose with type-specific rules (not all type combinations compose)
• The graph grows but converges — further analysis produces diminishing revisions

9.2 What the methodology does NOT impose

No constraints on graph topology. The methodology can produce:

• Linear chains (realization sequences)
• Trees (decomposition hierarchies)
• DAGs (mixed edge types with no cycles)
• Cycles (feedback loops, self-reference through L3)
• Clusters (configuration fans)
• Arbitrary connected structures

No constraints on domain content. Any domain with decomposable structure can be analyzed. The methodology doesn't prescribe what domains exist or what they contain. It prescribes HOW to analyze them, not WHAT to analyze.

No constraints on edge types. The current set (realization, configuration, role-id, enrichment, decomposition, feedback, selection) is empirically discovered, not prescribed. New types can be added when new structural relationships are found. The methodology is OPEN to new edge types.

9.3 Where constraints come from

The constraints in the methodology's graphs come from THREE sources:

1. Domain-internal constraints (Dp). Dependencies between primitives within a domain. These are discovered during L1 analysis and filter the lattice. They're STRUCTURAL constraints — imposed by the domain's own internal logic.

2. Cross-domain constraints (Cn). How positions in connected lattices restrict each other. These are discovered during L2 analysis. They're RELATIONAL constraints — imposed by the edge's content.

3. Analytical constraints (from L3 feedback). Patterns discovered at L3 create EXPECTATIONS for new analyses. These aren't hard constraints — they're predictions that can be refuted. They're SOFT constraints — the methodology's own learning modifying future analyses.

9.4 Can you encode arbitrary structure in edges?

Yes, in principle. Edge substance (Sb) is open-ended — it can carry whatever structural content the edge requires. The methodology doesn't restrict what substance can be. The CHARACTER (Ch) classification constrains what substance to EXPECT, but even character is an open set.

However: arbitrary encoding without structural analysis is POSSIBLE but NOT USEFUL within the methodology. The methodology's value comes from the structural analysis — identifying primitives, dependencies, interactions, compositions. If you just stuff arbitrary data into an edge's substance without analyzing it, you have an edge at Sb1 (substance identified but not structured). The methodology can accommodate it, but it's analytically thin.

9.5 What does the methodology actually constrain?

The methodology constrains the ANALYTICAL DEPTH, not the TOPOLOGY:

• Any topology is possible (chains, trees, DAGs, cycles, arbitrary)
• Any domain is analyzable (if it has decomposable structure)
• Any edge type is possible (the set is empirically extensible)
• But the DEPTH of analysis is constrained by the three-test methodology (primitives must be minimal, productive, recurrent; pairs must be classified; compositions must be irreducible)

The methodology is a FRAMEWORK FOR DEPTH, not a CONSTRAINT ON BREADTH. You can analyze anything — the methodology constrains how RIGOROUSLY you analyze it, not what you're allowed to analyze.

10. What does this tell us about the methodology?

10.1 The inter-layer graph is simple

Three nodes (L1, L2, L3) with a handful of typed edges. The graph isn't complex — it's a production chain with feedback:

L1 → L2 → L3 → (feedback) → L1

Plus embedding (L1 inside L2's Sb) and shortcuts (L1 directly to L3).

10.2 L2 is structurally necessary

You can go L1 → L3 directly (compare domains without formal edge analysis). But L2 adds substantial value:

• Bridge analysis reveals translation machinery (substance)
• Mapping reveals structural correspondences
• Constraints reveal cross-lattice restrictions
• Without L2, L3 can compare but can't EXPLAIN why domains relate

L3's patterns are richer when L2 has been done. The invariant topology is visible through L3 alone, but the EXPLANATION (bridge primitives with specific structure) requires L2.

10.3 The methodology doesn't constrain topology — it constrains depth

The methodology can produce any graph shape. It constrains HOW DEEPLY you analyze each node and edge, not what nodes and edges exist. The three-test primitive identification, the pair classification, the composition search — these are depth constraints, not breadth constraints.

This means the methodology is genuinely general — it can analyze any domain with decomposable structure and connect any analyzed domains through typed edges. The constraint is analytical rigor, not structural scope.

10.4 Edge types are empirically extensible

The current set of 7 edge types (realization, configuration, role-id, enrichment, decomposition, feedback, selection) is not closed. We may have discovered two more candidates in this analysis:

• Production — A produces what B consumes (process composition)
• Embedding — A's vocabulary is active inside B's primitives (contained analysis)

Whether these are genuinely new types or variants of existing ones needs more analysis. Production might be just process dependency. Embedding might be a variant of decomposition at the primitive level.

10.5 The full edge type question is open

We don't know if the current edge type set is complete. We don't know if the types form a closed or open set. We don't know if there are structural constraints on what KINDS of edges are possible. The methodology is empirically extensible — new types are added when found — but we haven't established any theory of what types MUST or CAN'T exist.

This is an open question for the methodology: is there a finite set of structurally distinct edge types, or is the set open-ended?

11. Summary

Edges between layers

Graph constraints

The methodology constrains ANALYTICAL DEPTH (how rigorously you analyze), not STRUCTURAL BREADTH (what you can analyze or what graph shapes you can produce). Any topology is possible. Any domain is analyzable. Edge types are empirically extensible. The methodology is a framework for depth applied to arbitrary structural breadth.

Open questions

1. Is the edge type set finite or open-ended? Is there a theory of what types are possible?
2. Is "production" a genuine edge type or just process dependency?
3. Is "embedding" a variant of decomposition or a distinct type?
4. What determines whether L2 is needed vs optional for a given L3 analysis?