Multi Dimensional Taxonomic Mapping for Complex Systems

20 concepts24 papers1 proposal

When analyzing complex systems—whether an emerging software ecosystem, a decentralized financial protocol, or a fragmented body of scientific literature—traditional, flat classification schemes frequently fail. Monolithic categorization collapses multi-faceted realities into single labels, obscuring hidden dependencies, systemic vulnerabilities, and structural gaps. For instance, labeling a security flaw simply as an "LLM exploit" fails to distinguish whether the vulnerability originates in the user prompt, the network protocol, or the host operating system.

To overcome this, researchers across diverse domains utilize multi-dimensional taxonomic mapping and layered architectural classification. By decomposing a system into orthogonal axes (dimensions) or vertical tiers (layers), this approach isolates variables, maps interactions, and uncovers blind spots that traditional analysis misses.


Layered Architectural Deconstruction: Isolating the "Where"

The core intuition behind a layered taxonomy is that complex systems are built out of decoupled, hierarchical tiers. When evaluating security or system behavior, mapping phenomena to specific architectural layers allows researchers to isolate risks and align interventions with the precise technical capabilities of each tier.

+-------------------------------------------------------+
| Host Environment / Physical Layer                     |
+-------------------------------------------------------+
                           ▲
                           │ (Inter-layer interfaces)
                           ▼
+-------------------------------------------------------+
| Communication Protocol / Network Plane                |
+-------------------------------------------------------+
                           ▲
                           │
                           ▼
+-------------------------------------------------------+
| Client Interface / Application Logic                  |
+-------------------------------------------------------+

A clear example of this is found in the security evaluation of agentic AI systems. In their benchmark of the Model Context Protocol (MCP), Yang et al. demonstrate that evaluating LLM agents purely through the lens of prompt injection is insufficient. Instead, the authors construct a four-layer framework consisting of the client interface, the communication protocol, the external tool server, and the host environment. This layered abstraction reveals that while client-side prompts might be secure, protocol-layer transport exploits can achieve a 100% success rate, exposing host-side sandboxes to escapes.

This structural isolation operates similarly across other highly technical domains:

  • Decentralized Finance (DeFi): Feng et al. (and their parallel work [Ding et al.]) show that standard smart contract audits suffer from a "transparency gap" because they treat code monolithically. By introducing a layered taxonomy that separates smart contract code, protocol logic, and third-party dependencies, the authors clarify why projects with flawless code execution still collapse due to high-level economic logic flaws.
  • Software-Defined Networking (SDN): Because SDN centralizes network control, it introduces a consolidated attack surface. Wu et al. apply the STRIDE threat model across a multi-plane taxonomy, separating the application, control, and data planes, alongside their specialized interfaces. This prevents the conflation of data-routing bugs with control-logic vulnerabilities.
  • Internet of Things (IoT) and Virtual Reality (VR): Choudhary et al. map IoT vulnerabilities directly to the perception, network, and processing layers to align security protocols with constrained hardware capabilities. Similarly, in virtual reality systems, Munilla Garrido et al. organize data privacy threats across hardware, client, server, and user layers, mapping defenses directly to specific stages of the data lifecycle to expose layers lacking adequate privacy controls.

Even in broader threat modeling, this intuition holds true. Hamid organizes digital stalking vectors into a multi-level hierarchy ranging from low-level technical breaches (like metadata exploitation) to high-level reconnaissance, allowing organizations to prioritize responses based on the attacker's entry point. In network defense, Shi et al. establish a hierarchical representation of network hosts based on asset importance, dynamically shifting defenses to protect high-value layers.


Orthogonal Facet Mapping: Deconstructing the Design Space

While layered classification handles vertical hierarchies, orthogonal facet mapping addresses systems where multiple independent variables interact simultaneously. Instead of asking where an issue occurs, multi-dimensional mapping asks what it looks like, why it happened, and how it behaves across independent axes.

Consider the challenge of diagnosing software compatibility in complex environments like mobile gaming. Song et al. construct a multi-dimensional classification framework that categorizes bugs along three independent axes: observable symptoms, underlying root causes, and remediation strategies.

                  [ Remediation Strategy ]
                         /
                        /
                       /
                      +------------------- [ Root Cause ]
                     /
                    /
                   /
         [ Observable Symptom ]

By decoupling these facets, the authors demonstrate that game compatibility issues are fundamentally driven by hardware fragmentation (such as GPU limitations and screen shapes) rather than software API changes—an insight that would be obscured if symptoms and root causes were lumped into a single classification list.

This multi-dimensional approach is equally powerful for mapping user needs and design spaces:

  • Creative Software Alignment: Zhao et al. employ a nine-dimensional taxonomy (including target population, usage context, and the role of automation) to map commercial writing tools against academic research and user discussions. By plotting these dimensions orthogonally, they expose a stark misalignment: commercial tools focus heavily on automated text generation, whereas writers are starved for tools that assist with non-linear organization and pre-writing tasks.
  • Human-AI Collaboration: To move past vague industry buzzwords, Holter maps the human-AI collaborative design space across three core pillars: agency (who commands), interaction (how communication flows), and adaptation (how learning occurs). Decomposing systems into these granular dimensions allows researchers to systematically compare disparate architectures.
  • System Configurations and Logic Flaws: Nabi et al. evaluate complex business logic vulnerabilities by cross-referencing internal security attributes, attacker patterns, flaw categories, and administrative scales. In the realm of physical security, Kalbo et al. categorize IP-based video surveillance systems by functional purpose, implementation style, and network topology, illustrating how specific architectural configurations directly yield unique cyber-physical vulnerabilities.

Systematic Meta-Analysis: Exposing Structural Gaps in Literature

When applied to meta-analysis, multi-dimensional taxonomic coding acts as a diagnostic tool to evaluate the maturity of entire research fields. When a research area grows rapidly, literature becomes fragmented. Coding papers across standardized, independent dimensions allows researchers to mathematically identify under-explored domains.

For example, Röttger et al. conducted a systematic review of 144 LLM safety datasets by coding them across dimensions such as creation method, language, and intended purpose. This structured approach transformed a chaotic landscape into a clear data map, exposing critical structural gaps—namely, a severe shortage of non-English datasets and a distinct disconnect between academic safety benchmarks and the evaluation practices actually used in major commercial model releases.

+------------------+-------------------+--------------------+--------------------+
| Paper / Dataset  | Dimension A       | Dimension B        | Dimension C        |
|                  | (e.g., Language)  | (e.g., Method)     | (e.g., Target)     |
+------------------+-------------------+--------------------+--------------------+
| Study 1          | English           | Synthetic          | Model Evaluation   |
| Study 2          | English           | Human-Curated      | Fine-Tuning        |
| Study 3          | Spanish           | Synthetic          | Model Evaluation   |
+------------------+-------------------+--------------------+--------------------+
| CRITICAL GAP:    | [ Non-English ]   | [ Human-Curated ]  | [ Fine-Tuning ]    |
+------------------+-------------------+--------------------+--------------------+

This systematic mapping reveals blind spots across other emerging fields:

  • Educational Technology: Mahon et al. analyzed the landscape of K-12 AI education by coding papers based on participant demographics, geographic location, and technical content. The multi-dimensional matrix immediately highlighted that research is heavily concentrated in the United States and focuses almost exclusively on students, leaving a glaring lack of data regarding teachers and the broader school ecosystem.
  • Machine Learning Visualization: Subramonyam et al. evaluated the generalizability of Visualization for Machine Learning (VIS4ML) research. By systematically coding papers across user expertise, task types, and model components, they proved an empirical "generalizability gap": most tools are evaluated against narrow, "solved" benchmark datasets rather than messy, real-world engineering workflows.
  • Open Data Trends: Enríquez-Reyes et al. mapped over a decade of open data research across technical topics, application domains, and process life-cycle phases, providing objective, empirical proof of how the field has transitioned from initial hype to practical exploitation by infomediaries.

Why the Idea Works

Multi-dimensional taxonomic mapping works because it replaces intuition with structural rigor. By breaking a complex system down into its constituent vertical layers or horizontal facets, it prevents distinct variables from confounding one another. Whether applied to the transport protocols of LLM agents, the root causes of mobile game crashes, or geographic biases in educational research, this methodology forces developers and researchers to see systems not as monolithic black boxes, but as structured spaces where gaps and vulnerabilities can be precisely isolated, measured, and mitigated.

Key Foundational Work

Online Harassment and Digital Stalking (2013) — contributed 1 concept elements to this direction.
An extensive systematic review on the Model-Driven Development of secure systems (2015) — contributed 1 concept elements to this direction.
A Systematic Review of Model-Driven Security (2013) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_multi_dimensional_literature_coding_6ad9c822,
  title = {multi-dimensional literature coding},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-dimensional-literature-coding/rdr_6ad9c8227a95409fbdef0eb0126ce1d8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_multi_dimensional_literature_coding_6ad9c822_v1,
  title = {multi-dimensional literature coding},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-dimensional-literature-coding/rdr_6ad9c8227a95409fbdef0eb0126ce1d8/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake