Interaction Graph Analysis¶
Relational graph model — instantiates Part-Level Explanatory Reduction
Maps which parts act on which as a network of nodes and interaction edges, so the relational structure behind a whole-level pattern becomes visible.
Reductionism explains wholes through parts and relations, and the Interaction Graph Analysis is the mechanism that supplies the relations. It renders the parts as nodes and draws an edge wherever one part acts on another — amplifies it, inhibits it, exchanges with it, constrains it — and labels each edge with the kind and strength of that interaction. The result is a topology: a map of the coupling structure that a bare inventory cannot show. Its defining move, and what separates it from every diagramming sibling, is that the graph is non-directional in intent: it captures who is coupled to whom across the whole population of parts, without committing to a single start-to-finish story. It answers "what is the shape of the interaction structure?" — not "what happened," and not "what is necessary."
Example¶
An ecologist is trying to explain why a lake's largest fish population crashed in a single season, a whole-level event the community keeps describing as "the lake went bad." Interaction Graph Analysis treats the lake's species and nutrient pools as nodes and draws the trophic edges between them: predation edges (bass eat minnows), competition edges (two zooplankton grazers), a nutrient-loading edge from farm runoff into the algae node, and a shading/oxygen edge from an algal bloom back onto the fish. Each edge carries a sign (+/−) and a rough weight.
Laid out as a graph, the structure tells a story no single row of an inventory could: the runoff node has an unusually high-weight positive edge into algae, algae has a strong negative edge (via oxygen depletion) onto the fish, and the fish node turns out to sit at a bottleneck — many paths run through it. The cross-level context edge from the neighboring farmland, external to the lake proper, is what closes the loop. The analysis has not proven the runoff caused the crash and has not tested whether removing any node would prevent it; it has revealed the relational skeleton — the coupling and the bottleneck — that makes the crash intelligible and tells the next mechanism exactly which edges are worth interrogating.
How it works¶
- Import the nodes. Take the part list as-is (typically from a Part Inventory Matrix) and treat each part as a node; the graph adds no new parts, only relations.
- Draw and type the edges. For every pair that interacts, draw an edge and label its kind — amplify, inhibit, exchange, constrain, substitute — and a sign and rough weight.
- Include cross-level context edges. Where an outside or higher-level factor acts on a part, draw it in as a boundary edge rather than pretending the parts are a closed system.
- Read the topology. Look for hubs, bottlenecks, feedback loops, and cut-points — structural features that flag where the whole-level pattern is likely organized — without asserting a direction of production.
Tuning parameters¶
- Edge inclusion threshold — how weak an interaction still earns an edge. A low threshold captures subtle couplings but produces an unreadable "hairball"; a high one is legible but may drop the decisive weak link.
- Directed vs. undirected — whether edges carry a direction of influence. Directed edges add causal nuance but tempt you toward a single-pathway reading that belongs to a different sibling.
- Edge typing richness — one generic "interacts" label versus a full vocabulary of interaction kinds. Richer types explain more but demand evidence for each.
- Weight resolution — binary present/absent versus graded weights. Weights enable centrality analysis but imply a precision the underlying evidence may not support.
- Context-edge scope — how far outside the part set you let boundary edges reach. Wider scope resists context-stripping but blurs the level boundary.
When it helps, and when it misleads¶
Its strength is exposing relational structure: hubs, feedback loops, and bottlenecks are properties of the organization of parts, invisible in any inventory, and they are frequently where the whole-level behavior is actually seated. When a team's disagreement is really about how the parts are wired together, the graph makes the wiring arguable.
Its failure mode is that a dense graph can be richly descriptive without being explanatory — a beautiful hairball[n1] that reifies mere co-occurrence as interaction and looks like understanding while testing nothing. The classic misuse is centrality theatre: computing that a node is "most connected" and treating that as proof it drives the whole, when high connectivity may be an artifact of measurement convenience or a spurious correlation drawn as an edge. The graph maps structure; it does not run the structure forward or knock any part out, so it cannot by itself establish that any edge is load-bearing. The guarding discipline is to treat the topology as a set of hypotheses about where explanation lives and route the promising edges to a mechanism that actually tests them.
How it implements the components¶
Interaction Graph Analysis fills the relational side of the archetype:
interaction_law_map— the edges, with their kinds, signs, and weights, are the interaction law map in graph form.constituent_part_inventory— the nodes carry the part inventory into a relational frame (imported rather than re-derived).cross_level_context_note— boundary edges record where outside or higher-level factors act on the parts, keeping context from being silently dropped.
It does NOT implement aggregation_bridge_rule or explanatory_sufficiency_claim — that's Mechanism Chain Diagram, its nearest twin: both draw parts and their couplings, but the Graph maps the whole non-directional topology of who-touches-whom, while the Chain selects one directed path and asserts it is sufficient to produce a specific outcome. It also leaves whole_level_reconstruction_test to the generative and test-side siblings.
Related¶
- Instantiates: Part-Level Explanatory Reduction — supplies the interaction structure the reduction needs to move beyond a list of parts.
- Consumes: Part Inventory Matrix — its parts become the graph's nodes.
- Sibling mechanisms: Part Inventory Matrix · Mechanism Chain Diagram · Bottom-Up Simulation · Ablation or Knockout Test · Aggregation Sensitivity Test · Residual Explanation Review · Scope Clause and Exception Note
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Interaction Graph Analysis operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it maps which parts act on which as a network of nodes and interaction edges, so the relational structure behind a whole-level pattern becomes visible
Independent corroboration: The frozen evidence defines Interaction Graph Analysis as 'Maps which parts act on which as a network of nodes and interaction edges, so the relational structure behind a whole-level pattern becomes visible', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Mathematics
Origin pattern: Convergent development
Present-day reach: Universal
Rationale: Representing entities as nodes and interactions as edges derives from graph theory.
Related originating lineages:
- Computer Science & Software Engineering — Graph algorithms and network-analysis tooling materially make structural patterns measurable at scale.
- Data Science & Analytics — Network analysis provides algorithms, empirical weighting, and visualization for observed interaction data.
- Systems Thinking & Cybernetics — Systems modeling materially interprets interaction topology as an explanation for whole-level behavior.
Review resolution: Both independent reviews place the primary lineage in mathematics. The queued differences (alternate_origin_disagreement) concern secondary metadata rather than primary provenance. The final retains computer_science, systems_cybernetics, data_science only where a reviewer supplied a formative-lineage rationale; this does not convert downstream applicability into origin. origin_mode=convergent because the reviewers document independently established or materially co-developing traditions. domain_reach=universal records application breadth separately from provenance.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] The "hairball" is the well-known critique of large network visualizations that are so densely connected they convey structure aesthetically but not analytically — every node touches every other, and no explanatory pattern can be read off the tangle. It is the graph-analysis analogue of the part-list masquerade: apparent completeness standing in for tested explanation. ↩