Cognitive Atlas 2.0: Personal Evidence Graphs, Backlinks, and Relationship Analysis

Full report

On this page

The transition of the Cognitive Atlas from a static, read-only map of derived object relationships into a genuinely useful, user-driven evidence and relationship research environment represents a fundamental shift in archival sense-making. The core operational directive of this expansion is to grant users the capability to generate, interrogate, and preserve structural hypotheses while enforcing strict boundaries between primary archival truths, system-derived inferences, and subjective user assertions. This capability is paramount within the Al.Qaeda.net Archival Reconstruction workstation, where the sheer volume of disparate intelligence, historical records, and fragmented communications requires robust tools to track analytical provenance. The environment must maintain the visual constraints of a late-1990s technical application. This aesthetic is characterized by rigid windowing systems, Multiple Document Interfaces (MDI), compact legends, tabbed property dialogs, and nested tree inspectors, purposefully eschewing modern, frictionless dashboard aesthetics in favor of high-information-density engineering tools. Concurrently, the underlying layout algorithms and performance optimizations must leverage contemporary computational strategies to ensure fluid interrogation of extensive knowledge graphs. Furthermore, a strict local-only requirement mandates that all subjective interpretations, spatial layouts, and custom ontologies remain securely on the user's local disk, permanently insulated from server-side ingestion or telemetry.

Historical and Modern Benchmark Analysis

To establish a foundational architecture for Cognitive Atlas 2.0, the analysis benchmarks the proposed features against a spectrum of historical and modern paradigms of network visualization, knowledge management, and data provenance. Understanding how precursor systems navigated the tension between automated layout and human cognitive modeling is essential for designing a tool that respects the analyst's mental map.

These precursor systems illustrate that successful investigative environments prioritize the structured recording of rationale. The derivation of insights is deeply tied to the ability to document the exact sequence of evidence evaluation. By synthesizing the visual density of 1990s CASE tools with the rigorous provenance tracking of genealogy software and modern ontological standards, Cognitive Atlas 2.0 can facilitate deep structural analysis without sacrificing data integrity.

Provenance Model and the Source-Integrity Rule

The highest priority constraint for Cognitive Atlas 2.0 is the absolute visual and structural segregation of data authority. An undifferentiated authority layer inevitably leads to analytical contamination, where a user's subjective hunch, once mapped visually, is later misinterpreted by that same user (or a collaborator viewing a shared portable workspace) as an established archival fact. The underlying provenance model adapts the concepts of the Open Provenance Model (OPM) and the World Wide Web Consortium's PROV-O specification. Specifically, the architecture must distinguish between the fixed Entities (the archival records), the Activities (the system's algorithmic extraction or the user's manual linking), and the Agents (the automated parser or the human researcher)15. In this paradigm, every node and edge acts as an assertion that must carry a cryptographic or structural signature denoting its origin.

Tri-Layer Authority Taxonomy

To enforce the source-integrity rule, all edges and nodes must be strictly typed and visually distinguishable according to a tri-layer authority taxonomy. These distinct visual rules cannot be overridden by user preference; they are hardcoded into the rendering engine.

Under no circumstances should a user-created node—a subjective grouping such as "Suspect Cell Alpha"—be stylistically confused with a source-backed entity, such as an official organization recognized in the archival record. The visual grammar must constantly broadcast the subjective, localized origin of user assertions13.

Relationship-Type Taxonomy and Edge Configuration

A nuanced relationship-type taxonomy requires the system to handle both the directional nature of assertions and the evidentiary weight supporting them. Nodes represent the nouns of the investigation, while edges represent the verbs and the evidence.

Directed vs. Undirected Links

The application must support three distinct edge topologies, each carrying specific semantic weight for downstream layout algorithms and pathfinding operations:

1. Directed (A → B): Indicates a one-way flow of command, communication, funding, or causality (e.g., Node A "funded" Node B; Node C "reports to" Node D).

2. Bi-directional (A ↔ B): Indicates confirmed mutual interaction, such as a two-way communication channel, a mutual financial transaction, or reciprocal affiliation.

3. Undirected (A — B): Represents simple association or correlation where causality or directionality is unknown, unproven, or irrelevant (e.g., Node A "was located near" Node B; Node C "is a sibling of" Node D).

“Related Records” Versus User Assertions

A critical distinction in the taxonomy is the separation between "Related Records" and user-driven assertions. "Related Records" are a function of the system's backend—when a user selects an archival document, the system may present a list of related records based on shared metadata, matching dates, or semantic similarity. These are System-Derived relationships. Conversely, a User Assertion is a deliberate, manual intervention where the analyst draws a line between two records and explicitly defines the nature of that relationship. While the system might suggest that Document A and Document B are related due to shared keywords, the user asserts that Document A contradicts Document B, or that Document A is a translation of Document B. The user's assertion acts as an overlay on top of the system's graph, entirely local and independent.

The Evidence Workflow: Citations and Confidence

Drawing inspiration from the GEDCOM genealogy standard, a single edge in Cognitive Atlas 2.0 is not merely a geometric line connecting two vertices; it is a discrete, complex data container9. A user asserting that "Person A is an alias of Person B" is making a substantive claim that requires evidentiary backing. The GEDCOM standard's SOURCE\_CITATION structure provides a perfect model, allowing multiple, distinct sources to support or refute a single lineage claim12.

The Edge Creation and Editing Process

The edge creation process invokes a classic late-1990s modal property dialog box. The dialog features bevels, a gray background, and a standard tabbed interface, avoiding any modern floating pop-overs. Tab 1: General Properties This tab defines the fundamental topology. It includes non-editable fields identifying the Source Node and Target Node. Dropdown menus allow the user to select the Link Type (e.g., Ownership, Communication, Alias, Proximity) based on a controlled vocabulary, with an option for "Custom Label." Radio buttons dictate the directionality of the edge. Tab 2: Evidence and Citations This tab is the core of the research environment. It contains a data grid (list view) allowing the analyst to attach multiple specific archival records to the single edge.

• An analyst can attach Document 104, tagging it as "Supporting Evidence."
• The same analyst can attach Document 212, tagging it as "Contradictory Evidence." Each citation entry permits a page or paragraph reference, mirroring the GEDCOM standard where a citation points not just to a book, but to a specific location within the source12.

Tab 3: Relationship Notes A standard, multi-line text area allowing the analyst to write a free-text justification for the link. This is essential for recording the visual analytic activity and the rationale by which an insight was derived, satisfying the requirements for insight provenance3. Tab 4: Confidence and Evidence Status A discrete radio button group (e.g., High, Medium, Low, Speculative) indicating the analyst's certainty in the relationship. This metric directly modulates the visual weight (thickness or opacity) of the edge on the canvas. If an edge has zero citations attached, it is flagged by the system as "Unsubstantiated," and its visual representation is rendered translucent or explicitly marked with a warning glyph.

Graph UI Layout and the Late-1990s Aesthetic

The visual constraint strictly forbids modern, polished network-analysis dashboards characterized by borderless windows, kinetic animations, and minimalist iconography. Instead, Cognitive Atlas 2.0 must emulate the classic, highly structured interface patterns of comprehensive analytical software from the 1996-2001 era.

Core Interface Components

The workspace utilizes a Multiple Document Interface (MDI). The main application frame features a standard gray background holding multiple draggable, resizable child windows. This allows a user to maintain a zoomed-out global graph in one sub-window while interrogating a tightly focused neighborhood view in another.

• Tool Palettes: A draggable, floating tool palette provides discrete cursor modes: Select, Pan, Insert Personal Node, Draw Directed Edge, Draw Undirected Edge, and Inspector Tool. Icons are strictly 16x16 or 32x32 pixel raster images with high contrast, lacking anti-aliasing.
• Tree/List Inspector: Docked to the left side of the screen is a standard hierarchical tree view (visually resembling Windows 98 Explorer). It acts as a dual-purpose outliner, showing all nodes currently present in the active graph workspace. Nodes are organized hierarchically by Entity Type (People, Locations, Documents, User Assertions) and sub-categorized by user-defined Collections or Cases.
• Status Bar: Located at the bottom of the main application window, it displays precise telemetry: node and edge counts, current zoom level, and readouts of the currently selected object (e.g., "Selected: Edge \[User Created\] \- High Confidence \- 2 Citations").
• Compact Legends: A floating, toggleable tool window acts as a legend, dynamically updating based on the node types and edge colors currently present in the active view, reinforcing the tri-layer authority taxonomy.

Exact Menus, Toolbars, and Context Operations

To fully define the late-1990s technical visualization convention, the interface relies on deep, structured menu bars and context-sensitive right-click menus.

Context operations (right-clicking a node on the canvas) immediately present a condensed version of the Node and View menus, allowing rapid access to "Expand Neighborhood," "Edit Properties," and "Find Path to Target."

Cross-Workstation Workflows and Integration

The integration of Cognitive Atlas 2.0 with the broader Al.Qaeda.net Archival Reconstruction workstation requires fluid, frictionless transitions from reading and searching into active graph building.

Workflows Starting in Explorer, Search, Reader, and Trail

1. From the Reader/Explorer: When a user is reading a digitized archival document and identifies two entities of interest (e.g., a person's name and a safehouse location), they can highlight the text, right-click, and select "Send to Cognitive Atlas." This action immediately generates nodes for the entities in the active workspace and attaches the source document as a citation to a prospective edge, awaiting the user to define the link type.

2. From the Search Interface: A user executing a complex Boolean search yielding 50 documents can highlight the results list and select "Generate Neighborhood Graph." The Atlas will spawn a new workspace, placing the 50 documents as nodes and querying the local database to draw any pre-existing archive-backed or system-derived edges between them.

3. From the Trail: The user's chronological search history or viewing trail can be bulk-exported into a timeline graph view. This establishes nodes for each document visited, linked sequentially by time, allowing the analyst to visualize their own investigative path.

The Backlink Architecture

A vital mechanism for serendipitous discovery is the backlink. Traditional graph software is often isolated from the text-reading experience. In Cognitive Atlas 2.0, whenever a user creates an edge between a user-created node (e.g., a theoretical construct like "Suspect Cell Alpha") and an archival document, the system writes a local backlink to that document's metadata3. If the user later encounters this archival document in the standard Reader view, a "Local Backlinks" status indicator in the Reader's margin alerts them that they have previously utilized this document in a specific saved graph view. Clicking the indicator reveals the context: "Linked to Suspect Cell Alpha in workspace Financial\_Tracking.caw." This architecture closes the loop between reading and graphing, ensuring that analytical labor is constantly recycled and surfaced during future reading sessions.

Analytical Graph Operations for Archive Research

To surpass mere visual mapping, the environment must facilitate rigorous analytical operations tailored for intelligence, law enforcement, and historical research.

Graph Search and Neighborhood Expansion

Locating specific nodes within dense networks is a primary challenge in graph visualization21.

• Search within Graph: A standard Ctrl+F invokes a modal dialog allowing the user to query node labels, edge notes, personal labels, or linked citations within the active workspace. Matching nodes are highlighted in bright yellow, and the viewport automatically centers on the bounding box of the results.
• Expand Neighborhood: A researcher often starts with a single "seed" node (e.g., a person of interest) and needs to discover their network. Right-clicking the node and selecting "Expand Neighborhood" prompts a dialog asking for the degree of separation (1-Degree or 2-Degree). The system queries the local database and dynamically injects all connected nodes up to the specified degree into the current canvas4. The layout algorithm then runs a localized relaxation pass to accommodate the new nodes without destroying the existing spatial map.
• Path Exploration: The "Find Path" tool allows an analyst to determine if and how two seemingly disparate entities are connected. The user selects a Source node and a Target node. The system utilizes shortest-path algorithms to calculate routes traversing archive-backed, derived, and user-created edges5. The resulting path (or top three alternative paths) is highlighted with a thick, glowing stroke, while all non-relevant nodes in the graph are temporarily dimmed (opacity reduced to 20%).

Matrix Analysis, Comparison, and ACH

Integrating methodologies from the intelligence community, specifically Richards Heuer's Analysis of Competing Hypotheses (ACH), the graph environment must interoperate with a matrix view22. While node-link diagrams are excellent for visualizing topology, matrices are superior for evaluating the weight of evidence against competing theories. Users can select a group of evidence nodes (documents) and a group of user-created assertion nodes (competing hypotheses) and invoke the "ACH Comparison Integration." This generates a grid window. The matrix automatically tallies the user-defined edge confidences (consistent vs. inconsistent evidence) connecting the documents to the hypotheses, allowing the analyst to evaluate which hypothesis holds the most evidentiary weight in a strict, tabular format24.

Filters, Saved Views, and Graph Search

Large datasets rapidly induce visual clutter, often referred to as a "hairball," rendering graphs unreadable and cognitively overwhelming21. Granular filtering is a mandatory capability to maintain the utility of the workspace.

Multi-Dimensional Filters

A dedicated "Graph Filters" tool window, dockable on the right side of the MDI, presents a series of checkboxes, dropdowns, and sliders to prune the visible graph dynamically.

• Source Class: Filter nodes by Archival Record, System Derived, or User Created. This allows an analyst to instantly strip away all subjective assertions and view only the hard, archive-backed facts17.
• Metadata Attributes: Filter nodes by book, research category, or specific user-defined collections.
• Temporal Filters (Date): A dual-slider timeline allowing users to filter out nodes or edges that possess chronological metadata outside the selected window. As the sliders are dragged, nodes appear and disappear, effectively animating the evolution of a network over time.
• Note / Evidence Status: Checkboxes allow the user to hide edges that lack user annotations or lack supporting citations, instantly revealing unsubstantiated claims within the graph.

Saved Views and Portable Workspaces

The spatial layout of a graph carries deep cognitive meaning for the user. When an analyst manually groups nodes on the screen, that spatial arrangement is an implicit assertion of relation, even if no edge is drawn26. Therefore, saving the workspace must record the exact [Figure omitted from source export] coordinates for all nodes. A "Saved View" operates as a spatial snapshot. A single underlying dataset within a project can have multiple Saved Views (e.g., "View 1: Financial Links," "View 2: Geographic Proximity"). This allows the user to pivot between different analytical contexts, applying different layouts or manual arrangements to the exact same set of nodes without losing their carefully curated mental map in the previous view.

Layout Paradigms, Aesthetics, and Performance

While the UI is styled for the late 1990s, the algorithmic positioning of nodes must rely on a rich history of graph drawing aesthetics and modern computational efficiencies to remain performant.

Algorithmic Layout Choices

Empirical studies on human comprehension of node-link diagrams demonstrate that users consistently prioritize the minimization of edge crossings, the minimization of edge bends, and the alignment of nodes to underlying grids to maximize readability13. To support diverse analytical tasks, the system provides three primary layout algorithms:

1. Sugiyama Framework (Hierarchical Layout): For directed acyclic graphs (DAGs), such as strict chains of command, reporting structures, or chronological event sequences, the system utilizes the Sugiyama algorithm. This four-step framework consists of: (1) cycle removal to ensure the graph flows in one direction, (2) layer assignment to place nodes into distinct horizontal or vertical strata, (3) crossing reduction via vertex permutation heuristics, and (4) coordinate assignment28. This layout ensures that directional flow is immediately visually apparent.

2. Force-Directed (Spring) Layout: For highly interconnected, undirected networks (e.g., social networks, communication webs), a force-directed model simulates physical repulsion between all nodes and spring attraction along edges. To handle large localized graphs efficiently, this layout relies on the Barnes-Hut approximation, which reduces the computational complexity of [Figure omitted from source export]\-body repulsion from [Figure omitted from source export] to [Figure omitted from source export] by recursively grouping distant nodes via a quadtree32.

3. Orthogonal Layout: Nodes are aligned to a strict invisible grid, and edges are drawn using purely horizontal and vertical segments with 90-degree bends. This aligns heavily with the 1990s CASE tool and UML diagram aesthetics, highly preferred by users for schematic or architectural data14.

User-Created Aesthetics and Layout Persistence

Algorithms provide a starting point, but manual layout is critical for investigative sense-making. Cognitive layout research indicates that the aesthetics favored by users during the active creation of a graph often differ from algorithmic optimizations. Users frequently group nodes logically (e.g., drawing implicit convex hulls around perceived clusters or placing high-value targets centrally) regardless of the mathematical edge topology26. Consequently, the system respects the user's manual interventions. Any manual repositioning of a node automatically locks its [Figure omitted from source export] coordinate, overriding subsequent automated layout sweeps unless the user explicitly executes a "Refresh Layout" command on that specific node. This ensures the absolute persistence of the user's localized spatial mapping26.

Large-Graph Performance Limits

While the aesthetic remains period-accurate, rendering thousands of nodes using actual 1990s code architectures would result in catastrophic UI freezing. Beneath the period GUI wrappers, the rendering engine must utilize modern optimizations. The graph visualization is implemented utilizing an HTML5 \<canvas\> element operating on an OffscreenCanvas via Web Workers. This architecture decouples the heavy physics calculations of the Barnes-Hut layout and the frame rendering from the main UI thread. As a result, the application's toolbars, menus, and property dialogs remain highly responsive even as a massive graph continuously stabilizes in the background34. Despite these computational optimizations (capable of handling 50,000+ nodes), the cognitive limits of human visual comprehension and the pixel limitations of rendering distinct 16x16 icons with legible text labels dictate a practical soft cap21. If a user attempts to render a neighborhood exceeding 2,500 nodes, the system triggers a warning dialog, advising the application of strict filters or suggesting a transition to the Tree/List view to prevent visual and cognitive overload.

Accessibility and Keyboard Navigation

Node-link diagrams are inherently hostile to visually impaired users or analysts who rely entirely on keyboard navigation for rapid data entry. Providing an accessible non-graph list/tree equivalent is not a secondary feature; it is an architectural necessity.

The WAI-ARIA TreeGrid Equivalent

To synthesize hierarchical navigation, the active two-dimensional graph canvas is simultaneously represented in the background (and visibly toggleable via the View menu) as a treegrid data structure, fully compliant with W3C WAI-ARIA specifications38.

• Structural Translation: A highly connected graph cannot be perfectly mapped to a standard hierarchical tree due to cyclical relationships (e.g., Node A links to B, B links to C, C links to A). To resolve this for the linear treegrid, the system dynamically extracts a Minimum Spanning Tree (MST) from the graph, prioritizing heavily weighted, high-confidence, or explicitly directed edges to establish parent-child relationships40. Remaining lateral edges (the cycles) are listed as cross-reference attributes in a separate column within the grid, ensuring no data is lost in the translation41.
• Navigation and Keyboard Operation: Keyboard users can utilize the Tab key to move focus between the tool palettes, the inspector, and the main canvas (represented to assistive technologies as the treegrid). Using the Up and Down arrow keys, they navigate between nodes (represented as rows). The Right arrow key expands a node to reveal its connected edges (children), and the Left arrow key collapses it41.

Sonification and Screen Readers

As focus moves to a node within the treegrid, standard ARIA properties announce the node's label, its entity type, its relationship count, and its source-authority level. aria-describedby tags are utilized to read the user's detailed notes associated with the node41. If the user navigates into an edge row, the screen reader announces the edge direction, the target node, and the evidence confidence level45. For sighted power-users operating the visual graph canvas, keyboard shortcuts are mapped to standard interactions to accelerate workflows without relying on a mouse: Arrow Keys to pan the canvas, Shift \+ Arrow Keys to move selected nodes in grid increments, Ctrl \+ A to select all, and the Spacebar to pulse the active layout algorithm42.

Data Portability: Local-Only Import/Export and Validation

Because all personal nodes, edges, labels, layouts, and graph views must remain strictly local to the user's workstation, data portability between isolated, air-gapped terminals (or for offline backup) relies on robust import and export mechanics. The system never syncs user graphs to a centralized server.

Export Formats

The environment supports four distinct export formats to accommodate different analytical needs:

1. Portable Workspace (.caw): A proprietary, compressed archive file containing the graph topology, all user notes, coordinate data for saved views, and UUID references to the archival documents. This allows a user to back up their exact working environment and load it flawlessly on another terminal.

2. JSON / GraphML: Standard machine-readable formats allowing the topological data (nodes, edges, attributes) to be exported for ingestion into third-party statistical network analysis software (e.g., Gephi or Python networkx scripts).

3. CSV Edge-List: A flat text file consisting of columns for Source, Target, Link\_Type, Confidence, Citation\_UUIDs, optimized for spreadsheet review, bulk editing, or algorithmic ingestion.

4. Human-Readable Relationship Report: A formatted text or HTML document that generates a narrative summary of the graph. It iterates through all User Assertions, listing the hypothesis, the connected entities, and printing the associated notes and citations in a standard bibliographic format, completely independent of the visual layout17.

Import Validation

When importing CSV edge-lists or Portable Workspaces, the system enforces a strict validation gate to maintain data integrity. The importer scans all UUIDs linking the imported edges to archival documents. If the target archival document is missing from the local database of the receiving workstation, the affected node is flagged with a red "\!" icon and labeled "Orphaned Reference." Furthermore, the import routine actively sanitizes the payload, stripping any external network URIs that might attempt to call out to remote servers for telemetry or image fetching, ensuring the hermetic seal of the local workstation is permanently maintained.

Risks of Misleading Visual Inference

The visual representation of data carries inherent cognitive risks that must be acknowledged and mitigated by the system's design. The foremost risk in graph visualization is driven by the Gestalt principle of proximity: human users instinctively assume that nodes placed close together on a screen belong to the same group or share characteristics, regardless of whether a line connects them26. In a force-directed layout, two highly connected clusters may be drawn near each other purely due to the mathematical constraints of the physics simulation attempting to fit the graph within the viewport, not because the clusters are conceptually related. Furthermore, the mapping of edge thickness (representing user confidence) and node size (representing degree centrality) can create a false sense of objective importance. A speculative user-created assertion drawn with a thick, high-confidence line can visually overpower a thin, unassailable archival fact. This visual imbalance can lead a subsequent viewer (or the same analyst returning weeks later) to mentally weigh the subjective hypothesis over the primary source, fundamentally compromising the integrity of the research. To mitigate these risks, Cognitive Atlas 2.0 must enforce the strict tri-layer styling rules (solid for facts, dashed for system, colored for user assertions) at the rendering level. Additionally, the system provides a persistent, floating "Legend" tool window that permanently reminds the user of the semantic mapping of shapes, colors, and line weights, ensuring the visual grammar is never misinterpreted.

Feature Priorities and Acceptance Criteria

To manage the implementation of this major expansion, development and testing must adhere to prioritized phases and strict acceptance criteria, ensuring foundational integrity before layering complex analytical tools.

By adhering to this comprehensive architectural and operational blueprint, Cognitive Atlas 2.0 will evolve from a passive visualization layer into a rigorous, local-only analytical environment. It will perfectly suit the meticulous demands of archival reconstruction, ensuring that human insight is captured, evidenced, and structurally preserved while honoring the visual and operational constraints of its technological era.

Works cited

1. March 2014 \- Electrospaces.net, https://www.electrospaces.net/2014/03/

2. Environmental Crime: Technical Investigation-Evidentiary/Forensic, https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=9100TI9W.TXT

3. Characterizing Users' Visual Analytic Activity for Insight Provenance, https://www.researchgate.net/publication/220041769\_Characterizing\_Users'\_Visual\_Analytic\_Activity\_for\_Insight\_Provenance

4. PajekManual | PDF, https://www.slideshare.net/slideshow/pajekmanual/6400201

5. Pajek: Network Analysis Software Guide | PDF | Graph Theory \- Scribd, https://www.scribd.com/document/145452240/Pajek-Analysis-Visualization-of-Large-Networks

6. Social Network Analysis : An Introduction with an Extensive, https://dokumen.pub/social-network-analysis-an-introduction-with-an-extensive-implementation-to-a-large-scale-online-network-using-pajek-1nbsped-9781608058181-9781608058198.html

7. Biologically Inspired Modular Neural Networks \- VTechWorks, https://vtechworks.lib.vt.edu/bitstreams/c6cd9bfa-d273-4fce-ae7a-7061d177e290/download

8. Middleware and Application Management Architecture, https://d-nb.info/965645207/34

9. GEDCOM \- Wikipedia, https://en.wikipedia.org/wiki/GEDCOM

10. The GEDCOM standard, https://user.it.uu.se/\~andersa/gedcom/intro.html

11. Research and evidence \- Charted Roots, https://chartedroots.com/research/

12. Standardizing Sources and Citation Templates, https://www.beholdgenealogy.com/blog/?p=1395

13. (PDF) Graph Drawing Aesthetics—Created by Users, Not Algorithms, https://www.researchgate.net/publication/49697910\_Graph\_Drawing\_Aesthetics-Created\_by\_Users\_Not\_Algorithms

14. Graph drawing aesthetics and the comprehension of UML class, https://www.researchgate.net/publication/2531384\_Graph\_drawing\_aesthetics\_and\_the\_comprehension\_of\_UML\_class\_diagrams\_an\_empirical\_study

15. PROV-O: The W3C Provenance Ontology \- CASRAI, https://casrai.org/dictionary/term/prov-o

16. Team Comment on the "The PROV-JSON Serialization" Submission, https://www.w3.org/submissions/2013/01/Comment/

17. The Foundations for Provenance on the Web | Emerald Publishing, https://www.emerald.com/ftweb/article/2/2-3/99/1323351/The-Foundations-for-Provenance-on-the-Web

18. PROV-O: The PROV Ontology \- Research Explorer, https://research.manchester.ac.uk/en/publications/prov-o-the-prov-ontology/

19. An Extensible Framework for Provenance in Human Terrain Visual, https://kaixu.me/wp-content/uploads/2013/08/diva\_vast\_2013\_revised.pdf

20. Citations and the GEDCOM \- Evidentia Software, https://evidentiasoftware.com/citations-and-the-gedccom/

21. Graph Visualization and Navigation in Information Visualization, https://www.ivan-herman.net/professional/Publications/StarGraphVisuInInfoVis.pdf

22. Intelligence Analysis for Tomorrow: Advances from the Behavioral, https://www.nationalacademies.org/read/13040/chapter/5

23. Psychology of Intelligence Analysis \- World Wars, https://www.allworldwars.com/Psychology-of-Intelligence-Analysis.html

24. Applying Heuer's Analysis of Competing Hypotheses (ACH) in the, https://www.academia.edu/38094272/Applying\_Heuers\_Analysis\_of\_Competing\_Hypotheses\_ACH\_in\_the\_Law\_Enforcement\_Intelligence\_Environment

25. Establishing Aesthetics based on Human Graph Reading Behavior, https://www.ieeesmc.org/wp-content/uploads/2015/09/tc-vac-paper.pdf

26. How Shape of Node-Link Diagrams Impacts Aesthetic Appreciation, https://fis.uni-bamberg.de/bitstreams/ccb23973-f17f-4c1a-96a3-9b70c7d2280c/download

27. Graph Drawing Aesthetics and the Comprehension of UML Class, https://www.semanticscholar.org/paper/Graph-Drawing-Aesthetics-and-the-Comprehension-of-Purchase-McGill/527ca0518fca9efdbea27c8a3289a4c8d67e22f6

28. Sugiyama Algorithm | PDF | Theoretical Computer Science \- Scribd, https://www.scribd.com/document/991255989/Sugiyama-Algorithm

29. (PDF) Sugiyama Algorithm \- ResearchGate, https://www.researchgate.net/publication/303226437\_Sugiyama\_Algorithm

30. Layered graph drawing \- Wikipedia, https://en.wikipedia.org/wiki/Layered\_graph\_drawing

31. Layered Graph Drawing for Visualizing Evaluation Structures, https://www.computer.org/csdl/magazine/cg/2017/02/mcg2017020020/13rRUB6SpUK

32. docker/php/app/node\_modules/d3-force \- GitLab, https://src.koda.cnrs.fr/aurelie.mace.1/findchem\_iscr/-/tree/25.01.16/docker/php/app/node\_modules/d3-force?ref\_type=tags

33. New 3D Graph \- Obsidian Plugin, https://community.obsidian.md/plugins/new-3d-graph

34. ForceParams | d3gl, https://mapequation.org/d3gl/reference/network/interfaces/forceparams/

35. A Comparison of User-Generated and Automatic Graph Layouts, https://petra.isenberg.cc/publications/papers/Dwyer\_2009\_ACO.pdf

36. Graph Drawing, https://teaching.vda.univie.ac.at/vis/14s/Presentations/140520\_Markovic.pdf

37. Speeding Up the Webcola Graph Viz Library with Rust and, https://news.ycombinator.com/item?id=27732136

38. Patterns | APG | WAI \- W3C, https://www.w3.org/WAI/ARIA/apg/patterns/

39. Treegrid Pattern | APG | WAI \- W3C, https://www.w3.org/WAI/ARIA/apg/patterns/treegrid/

40. Graph Visualization and Navigation in Information ... \- SlideServe, https://www.slideserve.com/garnett/graph-visualization-and-navigation-in-information-visualization-a-survey-powerpoint-ppt-presentation

41. Chart Reader: Accessible Visualization Experiences Designed with, https://alper.datav.is/publications/chartreader/

42. How to build accessible graph visualization tools, https://cambridge-intelligence.com/blog/build-accessible-data-visualization-apps-with-keylines/

43. ChartA11y: Designing Accessible Touch Experiences of ... \- arXiv, https://arxiv.org/html/2410.20545v1

44. The Ultimate Checklist for Accessible Data Visualisations, https://www.a11y-collective.com/blog/accessible-charts/

45. How to make interactive charts accessible \- Deque Systems, https://www.deque.com/blog/how-to-make-interactive-charts-accessible/